Changelog
What's new
What changed in the weekly run: new models, models that now meet the bar or fell below it, and changes in accuracy and price.
Get a weekly email when something changes
October 11, 2026
- New model GPT-6 Luna on Contract clauses · Czech: 100.00% correct fields. Model →
- New model DeepSeek V4.1 Flash on Contract clauses · Czech: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Contract clauses · Czech: 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Contract clauses · Czech: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Contract clauses · Czech: 100.00% correct fields. Model →
- New model Mistral Large 4 on Contract clauses · Czech: 100.00% correct fields. Model →
- New model GPT-6 Sol on Contract clauses · Czech: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Contract clauses · Czech: 100.00% correct fields. Model →
- New model Claude Sonnet 5 on Contract clauses · Czech: 100.00% correct fields. Model →
- New model Gemma 4 31B on Contract clauses · Czech: 99.95% correct fields. Model →
- New model Qwen3.8 Flash on Contract clauses · Czech: 99.90% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Contract clauses · Czech: 99.85% correct fields. Model →
- New model Mistral Small 4 on Contract clauses · Czech: 99.03% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Contract clauses · Czech: $0.23 per 1,000 documents. Model →
- New model GPT-6 Luna on Contract clauses · German: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Contract clauses · German: 100.00% correct fields. Model →
- New model GPT-6 Sol on Contract clauses · German: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Contract clauses · German: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Contract clauses · German: 99.90% correct fields. Model →
- New model Mistral Large 4 on Contract clauses · German: 99.44% correct fields. Model →
- New model Claude Sonnet 5 on Contract clauses · German: 98.21% correct fields. Model →
- New model Qwen3.8 Flash on Contract clauses · German: 98.00% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Contract clauses · German: 98.00% correct fields. Model →
- New model Mistral Medium 3.5 on Contract clauses · German: 97.13% correct fields. Model →
- New model DeepSeek V4.1 Flash on Contract clauses · German: 97.08% correct fields. Model →
- New model Gemma 4 31B on Contract clauses · German: 94.41% correct fields. Model →
- New model Mistral Small 4 on Contract clauses · German: 94.26% correct fields. Model →
- GPT-6 Luna is now the cheapest model that meets the bar on Contract clauses · German: $0.27 per 1,000 documents. Model →
- New model DeepSeek V4.1 Flash on Contract clauses · English (US): 100.00% correct fields. Model →
- New model GPT-6 Sol on Contract clauses · English (US): 100.00% correct fields. Model →
- New model Claude Sonnet 5 on Contract clauses · English (US): 100.00% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Contract clauses · English (US): 99.95% correct fields. Model →
- New model Gemini 3.8 Flash on Contract clauses · English (US): 99.90% correct fields. Model →
- New model GPT-6 Luna on Contract clauses · English (US): 99.79% correct fields. Model →
- New model Claude Sonnet 5.5 on Contract clauses · English (US): 99.79% correct fields. Model →
- New model Gemma 4 31B on Contract clauses · English (US): 99.64% correct fields. Model →
- New model GLM-5.3 Flash on Contract clauses · English (US): 99.49% correct fields. Model →
- New model Mistral Medium 3.5 on Contract clauses · English (US): 98.10% correct fields. Model →
- New model Mistral Large 4 on Contract clauses · English (US): 96.10% correct fields. Model →
- New model Qwen3.8 Flash on Contract clauses · English (US): 95.13% correct fields. Model →
- New model Mistral Small 4 on Contract clauses · English (US): 91.54% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Contract clauses · English (US): $0.18 per 1,000 documents. Model →
- New model Mistral Small 4 on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model GPT-6 Luna on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model DeepSeek V4.1 Flash on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model Qwen3.8 Flash on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model GPT-6 Sol on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model Mistral Large 4 on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model Claude Sonnet 5 on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Contract clauses · Spanish: 99.79% correct fields. Model →
- New model Gemma 4 31B on Contract clauses · Spanish: 94.41% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Contract clauses · Spanish: $0.23 per 1,000 documents. Model →
- New model Mistral Medium 3.5 on Contract clauses · French: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Contract clauses · French: 100.00% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Contract clauses · French: 99.95% correct fields. Model →
- New model Claude Sonnet 5.5 on Contract clauses · French: 99.95% correct fields. Model →
- New model Claude Sonnet 5 on Contract clauses · French: 99.95% correct fields. Model →
- New model DeepSeek V4.1 Flash on Contract clauses · French: 99.85% correct fields. Model →
- New model GLM-5.3 Flash on Contract clauses · French: 99.74% correct fields. Model →
- New model GPT-6 Sol on Contract clauses · French: 99.59% correct fields. Model →
- New model Qwen3.8 Flash on Contract clauses · French: 99.49% correct fields. Model →
- New model GPT-6 Luna on Contract clauses · French: 99.33% correct fields. Model →
- New model Mistral Large 4 on Contract clauses · French: 99.28% correct fields. Model →
- New model Mistral Small 4 on Contract clauses · French: 99.23% correct fields. Model →
- New model Gemma 4 31B on Contract clauses · French: 94.62% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Contract clauses · French: $0.23 per 1,000 documents. Model →
- New model Gemini 3.8 Flash on Email triage · Czech: 99.19% correct fields. Model →
- New model GPT-6 Sol on Email triage · Czech: 98.96% correct fields. Model →
- New model GLM-5.3 Flash on Email triage · Czech: 98.89% correct fields. Model →
- New model Claude Sonnet 5.5 on Email triage · Czech: 98.67% correct fields. Model →
- New model Qwen3.8 Flash on Email triage · Czech: 98.52% correct fields. Model →
- New model Mistral Large 4 on Email triage · Czech: 98.15% correct fields. Model →
- New model Claude Sonnet 5 on Email triage · Czech: 98.15% correct fields. Model →
- New model GPT-6 Luna on Email triage · Czech: 97.93% correct fields. Model →
- New model DeepSeek V4.1 Flash on Email triage · Czech: 97.41% correct fields. Model →
- New model Gemma 4 31B on Email triage · Czech: 97.11% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Email triage · Czech: 95.78% correct fields. Model →
- New model Mistral Medium 3.5 on Email triage · Czech: 94.52% correct fields. Model →
- New model Mistral Small 4 on Email triage · Czech: 87.63% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Email triage · Czech: $0.16 per 1,000 documents. Model →
- New model Gemini 3.8 Flash on Email triage · German: 98.37% correct fields. Model →
- New model GPT-6 Sol on Email triage · German: 98.22% correct fields. Model →
- New model Claude Sonnet 5.5 on Email triage · German: 98.00% correct fields. Model →
- New model GLM-5.3 Flash on Email triage · German: 98.00% correct fields. Model →
- New model Mistral Large 4 on Email triage · German: 97.85% correct fields. Model →
- New model Qwen3.8 Flash on Email triage · German: 97.85% correct fields. Model →
- New model Claude Sonnet 5 on Email triage · German: 97.85% correct fields. Model →
- New model GPT-6 Luna on Email triage · German: 96.96% correct fields. Model →
- New model DeepSeek V4.1 Flash on Email triage · German: 96.37% correct fields. Model →
- New model Gemma 4 31B on Email triage · German: 95.11% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Email triage · German: 94.67% correct fields. Model →
- New model Mistral Medium 3.5 on Email triage · German: 92.74% correct fields. Model →
- New model Mistral Small 4 on Email triage · German: 87.26% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Email triage · German: $0.15 per 1,000 documents. Model →
- New model GLM-5.3 Flash on Email triage · English (US): 99.41% correct fields. Model →
- New model GPT-6 Sol on Email triage · English (US): 99.04% correct fields. Model →
- New model Claude Sonnet 5.5 on Email triage · English (US): 98.89% correct fields. Model →
- New model Gemini 3.8 Flash on Email triage · English (US): 98.74% correct fields. Model →
- New model Mistral Large 4 on Email triage · English (US): 98.74% correct fields. Model →
- New model Claude Sonnet 5 on Email triage · English (US): 98.22% correct fields. Model →
- New model DeepSeek V4.1 Flash on Email triage · English (US): 97.63% correct fields. Model →
- New model Qwen3.8 Flash on Email triage · English (US): 97.48% correct fields. Model →
- New model GPT-6 Luna on Email triage · English (US): 97.48% correct fields. Model →
- New model Gemma 4 31B on Email triage · English (US): 96.52% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Email triage · English (US): 96.30% correct fields. Model →
- New model Mistral Medium 3.5 on Email triage · English (US): 94.89% correct fields. Model →
- New model Mistral Small 4 on Email triage · English (US): 89.70% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Email triage · English (US): $0.12 per 1,000 documents. Model →
- New model GLM-5.3 Flash on Email triage · Spanish: 99.11% correct fields. Model →
- New model Gemini 3.8 Flash on Email triage · Spanish: 98.96% correct fields. Model →
- New model GPT-6 Sol on Email triage · Spanish: 98.44% correct fields. Model →
- New model Mistral Large 4 on Email triage · Spanish: 97.93% correct fields. Model →
- New model Claude Sonnet 5.5 on Email triage · Spanish: 97.70% correct fields. Model →
- New model Claude Sonnet 5 on Email triage · Spanish: 97.56% correct fields. Model →
- New model Gemma 4 31B on Email triage · Spanish: 97.11% correct fields. Model →
- New model GPT-6 Luna on Email triage · Spanish: 96.81% correct fields. Model →
- New model Qwen3.8 Flash on Email triage · Spanish: 96.74% correct fields. Model →
- New model DeepSeek V4.1 Flash on Email triage · Spanish: 96.52% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Email triage · Spanish: 95.56% correct fields. Model →
- New model Mistral Medium 3.5 on Email triage · Spanish: 94.15% correct fields. Model →
- New model Mistral Small 4 on Email triage · Spanish: 88.67% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Email triage · Spanish: $0.15 per 1,000 documents. Model →
- New model Claude Sonnet 5 on Email triage · French: 98.67% correct fields. Model →
- New model Gemini 3.8 Flash on Email triage · French: 98.59% correct fields. Model →
- New model GPT-6 Sol on Email triage · French: 98.59% correct fields. Model →
- New model GLM-5.3 Flash on Email triage · French: 98.52% correct fields. Model →
- New model Mistral Large 4 on Email triage · French: 98.30% correct fields. Model →
- New model Claude Sonnet 5.5 on Email triage · French: 97.78% correct fields. Model →
- New model GPT-6 Luna on Email triage · French: 97.26% correct fields. Model →
- New model DeepSeek V4.1 Flash on Email triage · French: 96.74% correct fields. Model →
- New model Qwen3.8 Flash on Email triage · French: 96.44% correct fields. Model →
- New model Gemma 4 31B on Email triage · French: 96.22% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Email triage · French: 96.15% correct fields. Model →
- New model Mistral Medium 3.5 on Email triage · French: 94.15% correct fields. Model →
- New model Mistral Small 4 on Email triage · French: 90.52% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Email triage · French: $0.16 per 1,000 documents. Model →
- New model Claude Sonnet 5 on Invoice extraction · Czech: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Invoice extraction · Czech: 99.95% correct fields. Model →
- New model Claude Sonnet 5.5 on Invoice extraction · Czech: 99.95% correct fields. Model →
- New model DeepSeek V4.1 Flash on Invoice extraction · Czech: 99.90% correct fields. Model →
- New model GPT-6 Sol on Invoice extraction · Czech: 99.90% correct fields. Model →
- New model GLM-5.3 Flash on Invoice extraction · Czech: 99.69% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Invoice extraction · Czech: 99.49% correct fields. Model →
- New model GPT-6 Luna on Invoice extraction · Czech: 99.28% correct fields. Model →
- New model Mistral Large 4 on Invoice extraction · Czech: 97.08% correct fields. Model →
- New model Gemma 4 31B on Invoice extraction · Czech: 96.10% correct fields. Model →
- New model Mistral Small 4 on Invoice extraction · Czech: 93.64% correct fields. Model →
- New model Mistral Medium 3.5 on Invoice extraction · Czech: 93.49% correct fields. Model →
- New model Qwen3.8 Flash on Invoice extraction · Czech: 91.79% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Invoice extraction · Czech: $0.21 per 1,000 documents. Model →
- New model Gemini 3.5 Flash Lite on Invoice extraction · German: 100.00% correct fields. Model →
- New model GPT-6 Sol on Invoice extraction · German: 99.95% correct fields. Model →
- New model Claude Sonnet 5.5 on Invoice extraction · German: 99.95% correct fields. Model →
- New model GLM-5.3 Flash on Invoice extraction · German: 99.54% correct fields. Model →
- New model DeepSeek V4.1 Flash on Invoice extraction · German: 99.54% correct fields. Model →
- New model GPT-6 Luna on Invoice extraction · German: 99.23% correct fields. Model →
- New model Claude Sonnet 5 on Invoice extraction · German: 98.97% correct fields. Model →
- New model Mistral Large 4 on Invoice extraction · German: 97.49% correct fields. Model →
- New model Mistral Medium 3.5 on Invoice extraction · German: 97.38% correct fields. Model →
- New model Gemini 3.8 Flash on Invoice extraction · German: 97.23% correct fields. Model →
- New model Gemma 4 31B on Invoice extraction · German: 94.87% correct fields. Model →
- New model Mistral Small 4 on Invoice extraction · German: 94.82% correct fields. Model →
- New model Qwen3.8 Flash on Invoice extraction · German: 93.49% correct fields. Model →
- GPT-6 Luna is now the cheapest model that meets the bar on Invoice extraction · German: $0.54 per 1,000 documents. Model →
- New model Claude Sonnet 5.5 on Invoice extraction · English (US): 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Invoice extraction · English (US): 99.94% correct fields. Model →
- New model GPT-6 Sol on Invoice extraction · English (US): 99.94% correct fields. Model →
- New model Claude Sonnet 5 on Invoice extraction · English (US): 99.82% correct fields. Model →
- New model Gemini 3.8 Flash on Invoice extraction · English (US): 99.82% correct fields. Model →
- New model GPT-6 Luna on Invoice extraction · English (US): 99.70% correct fields. Model →
- New model DeepSeek V4.1 Flash on Invoice extraction · English (US): 99.39% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Invoice extraction · English (US): 99.21% correct fields. Model →
- New model Mistral Medium 3.5 on Invoice extraction · English (US): 97.76% correct fields. Model →
- New model Gemma 4 31B on Invoice extraction · English (US): 97.64% correct fields. Model →
- New model Mistral Small 4 on Invoice extraction · English (US): 95.58% correct fields. Model →
- New model Mistral Large 4 on Invoice extraction · English (US): 95.39% correct fields. Model →
- New model Qwen3.8 Flash on Invoice extraction · English (US): 82.36% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Invoice extraction · English (US): $0.17 per 1,000 documents. Model →
- New model Claude Sonnet 5.5 on Invoice extraction · Spanish: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Invoice extraction · Spanish: 99.38% correct fields. Model →
- New model GPT-6 Sol on Invoice extraction · Spanish: 99.23% correct fields. Model →
- New model GPT-6 Luna on Invoice extraction · Spanish: 99.13% correct fields. Model →
- New model Claude Sonnet 5 on Invoice extraction · Spanish: 98.46% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Invoice extraction · Spanish: 98.21% correct fields. Model →
- New model DeepSeek V4.1 Flash on Invoice extraction · Spanish: 98.21% correct fields. Model →
- New model Gemini 3.8 Flash on Invoice extraction · Spanish: 97.28% correct fields. Model →
- New model Mistral Medium 3.5 on Invoice extraction · Spanish: 96.87% correct fields. Model →
- New model Mistral Large 4 on Invoice extraction · Spanish: 96.51% correct fields. Model →
- New model Gemma 4 31B on Invoice extraction · Spanish: 94.62% correct fields. Model →
- New model Mistral Small 4 on Invoice extraction · Spanish: 92.36% correct fields. Model →
- New model Qwen3.8 Flash on Invoice extraction · Spanish: 88.87% correct fields. Model →
- GPT-6 Luna is now the cheapest model that meets the bar on Invoice extraction · Spanish: $0.55 per 1,000 documents. Model →
- New model Gemini 3.8 Flash on Invoice extraction · French: 99.95% correct fields. Model →
- New model Claude Sonnet 5.5 on Invoice extraction · French: 99.95% correct fields. Model →
- New model GPT-6 Sol on Invoice extraction · French: 99.90% correct fields. Model →
- New model Claude Sonnet 5 on Invoice extraction · French: 99.69% correct fields. Model →
- New model DeepSeek V4.1 Flash on Invoice extraction · French: 99.54% correct fields. Model →
- New model GLM-5.3 Flash on Invoice extraction · French: 99.08% correct fields. Model →
- New model GPT-6 Luna on Invoice extraction · French: 99.08% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Invoice extraction · French: 98.97% correct fields. Model →
- New model Mistral Medium 3.5 on Invoice extraction · French: 98.56% correct fields. Model →
- New model Mistral Small 4 on Invoice extraction · French: 94.87% correct fields. Model →
- New model Gemma 4 31B on Invoice extraction · French: 93.54% correct fields. Model →
- New model Qwen3.8 Flash on Invoice extraction · French: 90.46% correct fields. Model →
- New model Mistral Large 4 on Invoice extraction · French: 88.97% correct fields. Model →
- GPT-6 Luna is now the cheapest model that meets the bar on Invoice extraction · French: $0.56 per 1,000 documents. Model →
- New model Gemma 4 31B on Personal data detection · Czech: 99.81% correct fields. Model →
- New model Mistral Medium 3.5 on Personal data detection · Czech: 99.81% correct fields. Model →
- New model DeepSeek V4.1 Flash on Personal data detection · Czech: 99.71% correct fields. Model →
- New model GPT-6 Sol on Personal data detection · Czech: 99.71% correct fields. Model →
- New model Claude Sonnet 5.5 on Personal data detection · Czech: 99.71% correct fields. Model →
- New model GPT-6 Luna on Personal data detection · Czech: 99.62% correct fields. Model →
- New model Gemini 3.8 Flash on Personal data detection · Czech: 99.52% correct fields. Model →
- New model GLM-5.3 Flash on Personal data detection · Czech: 99.33% correct fields. Model →
- New model Mistral Large 4 on Personal data detection · Czech: 99.33% correct fields. Model →
- New model Claude Sonnet 5 on Personal data detection · Czech: 98.86% correct fields. Model →
- New model Mistral Small 4 on Personal data detection · Czech: 98.76% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Personal data detection · Czech: 98.10% correct fields. Model →
- New model Qwen3.8 Flash on Personal data detection · Czech: 96.95% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Personal data detection · Czech: $0.09 per 1,000 documents. Model →
- New model Mistral Medium 3.5 on Personal data detection · German: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Personal data detection · German: 100.00% correct fields. Model →
- New model Gemma 4 31B on Personal data detection · German: 99.90% correct fields. Model →
- New model DeepSeek V4.1 Flash on Personal data detection · German: 99.81% correct fields. Model →
- New model GLM-5.3 Flash on Personal data detection · German: 99.81% correct fields. Model →
- New model Gemini 3.8 Flash on Personal data detection · German: 99.71% correct fields. Model →
- New model GPT-6 Sol on Personal data detection · German: 99.71% correct fields. Model →
- New model Claude Sonnet 5 on Personal data detection · German: 99.71% correct fields. Model →
- New model Qwen3.8 Flash on Personal data detection · German: 99.62% correct fields. Model →
- New model GPT-6 Luna on Personal data detection · German: 99.62% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Personal data detection · German: 99.62% correct fields. Model →
- New model Mistral Large 4 on Personal data detection · German: 99.52% correct fields. Model →
- New model Mistral Small 4 on Personal data detection · German: 99.14% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Personal data detection · German: $0.09 per 1,000 documents. Model →
- New model GLM-5.3 Flash on Personal data detection · English (US): 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Personal data detection · English (US): 100.00% correct fields. Model →
- New model Gemma 4 31B on Personal data detection · English (US): 99.90% correct fields. Model →
- New model Mistral Medium 3.5 on Personal data detection · English (US): 99.71% correct fields. Model →
- New model Gemini 3.8 Flash on Personal data detection · English (US): 99.71% correct fields. Model →
- New model GPT-6 Sol on Personal data detection · English (US): 99.71% correct fields. Model →
- New model Claude Sonnet 5 on Personal data detection · English (US): 99.71% correct fields. Model →
- New model GPT-6 Luna on Personal data detection · English (US): 99.62% correct fields. Model →
- New model DeepSeek V4.1 Flash on Personal data detection · English (US): 99.62% correct fields. Model →
- New model Mistral Large 4 on Personal data detection · English (US): 99.62% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Personal data detection · English (US): 99.52% correct fields. Model →
- New model Mistral Small 4 on Personal data detection · English (US): 99.43% correct fields. Model →
- New model Qwen3.8 Flash on Personal data detection · English (US): 99.24% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Personal data detection · English (US): $0.08 per 1,000 documents. Model →
- New model Claude Sonnet 5.5 on Personal data detection · Spanish: 100.00% correct fields. Model →
- New model Gemma 4 31B on Personal data detection · Spanish: 99.90% correct fields. Model →
- New model DeepSeek V4.1 Flash on Personal data detection · Spanish: 99.90% correct fields. Model →
- New model Qwen3.8 Flash on Personal data detection · Spanish: 99.81% correct fields. Model →
- New model GLM-5.3 Flash on Personal data detection · Spanish: 99.81% correct fields. Model →
- New model Gemini 3.8 Flash on Personal data detection · Spanish: 99.81% correct fields. Model →
- New model GPT-6 Sol on Personal data detection · Spanish: 99.81% correct fields. Model →
- New model Mistral Large 4 on Personal data detection · Spanish: 99.81% correct fields. Model →
- New model Claude Sonnet 5 on Personal data detection · Spanish: 99.81% correct fields. Model →
- New model GPT-6 Luna on Personal data detection · Spanish: 99.71% correct fields. Model →
- New model Mistral Medium 3.5 on Personal data detection · Spanish: 99.62% correct fields. Model →
- New model Mistral Small 4 on Personal data detection · Spanish: 99.05% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Personal data detection · Spanish: 95.81% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Personal data detection · Spanish: $0.09 per 1,000 documents. Model →
- New model Claude Sonnet 5.5 on Personal data detection · French: 100.00% correct fields. Model →
- New model Gemma 4 31B on Personal data detection · French: 99.81% correct fields. Model →
- New model GLM-5.3 Flash on Personal data detection · French: 99.81% correct fields. Model →
- New model Gemini 3.8 Flash on Personal data detection · French: 99.81% correct fields. Model →
- New model DeepSeek V4.1 Flash on Personal data detection · French: 99.71% correct fields. Model →
- New model Qwen3.8 Flash on Personal data detection · French: 99.71% correct fields. Model →
- New model GPT-6 Sol on Personal data detection · French: 99.71% correct fields. Model →
- New model Claude Sonnet 5 on Personal data detection · French: 99.71% correct fields. Model →
- New model GPT-6 Luna on Personal data detection · French: 99.62% correct fields. Model →
- New model Mistral Medium 3.5 on Personal data detection · French: 99.62% correct fields. Model →
- New model Mistral Small 4 on Personal data detection · French: 98.95% correct fields. Model →
- New model Mistral Large 4 on Personal data detection · French: 98.67% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Personal data detection · French: 94.95% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Personal data detection · French: $0.09 per 1,000 documents. Model →
- New model Gemma 4 31B on Questions about a document · Czech: 100.00% correct fields. Model →
- New model Mistral Small 4 on Questions about a document · Czech: 100.00% correct fields. Model →
- New model GPT-6 Luna on Questions about a document · Czech: 100.00% correct fields. Model →
- New model DeepSeek V4.1 Flash on Questions about a document · Czech: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Questions about a document · Czech: 100.00% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Questions about a document · Czech: 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Questions about a document · Czech: 100.00% correct fields. Model →
- New model GPT-6 Sol on Questions about a document · Czech: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Questions about a document · Czech: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Questions about a document · Czech: 100.00% correct fields. Model →
- New model Claude Sonnet 5 on Questions about a document · Czech: 100.00% correct fields. Model →
- New model Qwen3.8 Flash on Questions about a document · Czech: 94.22% correct fields. Model →
- New model Mistral Large 4 on Questions about a document · Czech: 89.89% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Questions about a document · Czech: $0.21 per 1,000 documents. Model →
- New model Mistral Small 4 on Questions about a document · German: 100.00% correct fields. Model →
- New model Gemma 4 31B on Questions about a document · German: 100.00% correct fields. Model →
- New model DeepSeek V4.1 Flash on Questions about a document · German: 100.00% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Questions about a document · German: 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Questions about a document · German: 100.00% correct fields. Model →
- New model GPT-6 Sol on Questions about a document · German: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Questions about a document · German: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Questions about a document · German: 100.00% correct fields. Model →
- New model Claude Sonnet 5 on Questions about a document · German: 100.00% correct fields. Model →
- New model Qwen3.8 Flash on Questions about a document · German: 98.89% correct fields. Model →
- New model GLM-5.3 Flash on Questions about a document · German: 98.78% correct fields. Model →
- New model GPT-6 Luna on Questions about a document · German: 98.00% correct fields. Model →
- New model Mistral Large 4 on Questions about a document · German: 93.56% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Questions about a document · German: $0.18 per 1,000 documents. Model →
- New model Mistral Small 4 on Questions about a document · English (US): 100.00% correct fields. Model →
- New model Gemma 4 31B on Questions about a document · English (US): 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Questions about a document · English (US): 100.00% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Questions about a document · English (US): 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Questions about a document · English (US): 100.00% correct fields. Model →
- New model GPT-6 Sol on Questions about a document · English (US): 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Questions about a document · English (US): 100.00% correct fields. Model →
- New model Claude Sonnet 5 on Questions about a document · English (US): 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Questions about a document · English (US): 99.89% correct fields. Model →
- New model Qwen3.8 Flash on Questions about a document · English (US): 99.78% correct fields. Model →
- New model DeepSeek V4.1 Flash on Questions about a document · English (US): 99.22% correct fields. Model →
- New model GPT-6 Luna on Questions about a document · English (US): 99.11% correct fields. Model →
- New model Mistral Large 4 on Questions about a document · English (US): 98.89% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Questions about a document · English (US): $0.15 per 1,000 documents. Model →
- New model Gemma 4 31B on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model Mistral Small 4 on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model DeepSeek V4.1 Flash on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model GPT-6 Sol on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model Claude Sonnet 5 on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model GPT-6 Luna on Questions about a document · Spanish: 99.67% correct fields. Model →
- New model Qwen3.8 Flash on Questions about a document · Spanish: 99.67% correct fields. Model →
- New model Mistral Large 4 on Questions about a document · Spanish: 94.22% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Questions about a document · Spanish: $0.18 per 1,000 documents. Model →
- New model Mistral Small 4 on Questions about a document · French: 100.00% correct fields. Model →
- New model Gemma 4 31B on Questions about a document · French: 100.00% correct fields. Model →
- New model DeepSeek V4.1 Flash on Questions about a document · French: 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Questions about a document · French: 100.00% correct fields. Model →
- New model GPT-6 Sol on Questions about a document · French: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Questions about a document · French: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Questions about a document · French: 100.00% correct fields. Model →
- New model Claude Sonnet 5 on Questions about a document · French: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Questions about a document · French: 99.67% correct fields. Model →
- New model GPT-6 Luna on Questions about a document · French: 99.56% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Questions about a document · French: 99.56% correct fields. Model →
- New model Qwen3.8 Flash on Questions about a document · French: 98.78% correct fields. Model →
- New model Mistral Large 4 on Questions about a document · French: 92.33% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Questions about a document · French: $0.18 per 1,000 documents. Model →
- New model Gemma 4 31B on Document summaries · Czech: 100.00% correct fields. Model →
- New model GPT-6 Luna on Document summaries · Czech: 100.00% correct fields. Model →
- New model DeepSeek V4.1 Flash on Document summaries · Czech: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Document summaries · Czech: 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Document summaries · Czech: 100.00% correct fields. Model →
- New model GPT-6 Sol on Document summaries · Czech: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Document summaries · Czech: 100.00% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Document summaries · Czech: 99.56% correct fields. Model →
- New model Mistral Small 4 on Document summaries · Czech: 99.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Document summaries · Czech: 98.67% correct fields. Model →
- New model Qwen3.8 Flash on Document summaries · Czech: 95.44% correct fields. Model →
- New model Claude Sonnet 5 on Document summaries · Czech: 93.44% correct fields. Model →
- New model Mistral Large 4 on Document summaries · Czech: 83.33% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Document summaries · Czech: $0.19 per 1,000 documents. Model →
- New model Gemma 4 31B on Document summaries · German: 100.00% correct fields. Model →
- New model GPT-6 Luna on Document summaries · German: 100.00% correct fields. Model →
- New model DeepSeek V4.1 Flash on Document summaries · German: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Document summaries · German: 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Document summaries · German: 100.00% correct fields. Model →
- New model GPT-6 Sol on Document summaries · German: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Document summaries · German: 100.00% correct fields. Model →
- New model Mistral Small 4 on Document summaries · German: 99.89% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Document summaries · German: 99.89% correct fields. Model →
- New model Mistral Large 4 on Document summaries · German: 99.33% correct fields. Model →
- New model Claude Sonnet 5 on Document summaries · German: 99.11% correct fields. Model →
- New model Claude Sonnet 5.5 on Document summaries · German: 99.11% correct fields. Model →
- New model Qwen3.8 Flash on Document summaries · German: 95.33% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Document summaries · German: $0.16 per 1,000 documents. Model →
- New model Gemma 4 31B on Document summaries · English (US): 100.00% correct fields. Model →
- New model GPT-6 Luna on Document summaries · English (US): 100.00% correct fields. Model →
- New model DeepSeek V4.1 Flash on Document summaries · English (US): 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Document summaries · English (US): 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Document summaries · English (US): 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Document summaries · English (US): 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Document summaries · English (US): 100.00% correct fields. Model →
- New model Mistral Large 4 on Document summaries · English (US): 100.00% correct fields. Model →
- New model Qwen3.8 Flash on Document summaries · English (US): 99.89% correct fields. Model →
- New model Mistral Small 4 on Document summaries · English (US): 99.67% correct fields. Model →
- New model GPT-6 Sol on Document summaries · English (US): 99.56% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Document summaries · English (US): 99.22% correct fields. Model →
- New model Claude Sonnet 5 on Document summaries · English (US): 96.00% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Document summaries · English (US): $0.13 per 1,000 documents. Model →
- New model DeepSeek V4.1 Flash on Document summaries · Spanish: 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Document summaries · Spanish: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Document summaries · Spanish: 100.00% correct fields. Model →
- New model Gemma 4 31B on Document summaries · Spanish: 99.67% correct fields. Model →
- New model GPT-6 Sol on Document summaries · Spanish: 97.89% correct fields. Model →
- New model Mistral Small 4 on Document summaries · Spanish: 97.78% correct fields. Model →
- New model GPT-6 Luna on Document summaries · Spanish: 97.00% correct fields. Model →
- New model GLM-5.3 Flash on Document summaries · Spanish: 96.67% correct fields. Model →
- New model Claude Sonnet 5.5 on Document summaries · Spanish: 95.56% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Document summaries · Spanish: 95.33% correct fields. Model →
- New model Qwen3.8 Flash on Document summaries · Spanish: 94.44% correct fields. Model →
- New model Claude Sonnet 5 on Document summaries · Spanish: 94.22% correct fields. Model →
- New model Mistral Large 4 on Document summaries · Spanish: 91.33% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Document summaries · Spanish: $0.16 per 1,000 documents. Model →
- New model DeepSeek V4.1 Flash on Document summaries · French: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Document summaries · French: 100.00% correct fields. Model →
- New model Mistral Medium 3.5 on Document summaries · French: 100.00% correct fields. Model →
- New model Gemini 3.8 Flash on Document summaries · French: 100.00% correct fields. Model →
- New model Gemma 4 31B on Document summaries · French: 99.11% correct fields. Model →
- New model GPT-6 Luna on Document summaries · French: 98.78% correct fields. Model →
- New model Claude Sonnet 5.5 on Document summaries · French: 97.89% correct fields. Model →
- New model GPT-6 Sol on Document summaries · French: 97.22% correct fields. Model →
- New model Claude Sonnet 5 on Document summaries · French: 96.78% correct fields. Model →
- New model Qwen3.8 Flash on Document summaries · French: 96.33% correct fields. Model →
- New model Gemini 3.5 Flash Lite on Document summaries · French: 94.67% correct fields. Model →
- New model Mistral Small 4 on Document summaries · French: 93.22% correct fields. Model →
- New model Mistral Large 4 on Document summaries · French: 72.67% correct fields. Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Document summaries · French: $0.16 per 1,000 documents. Model →
October 10, 2026
- New model Claude Haiku 5.5 on Contract clauses · Czech: 100.00% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Contract clauses · Czech: $0.40 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Contract clauses · German: 100.00% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Contract clauses · German: $0.48 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Contract clauses · English (US): 99.69% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Contract clauses · English (US): $0.35 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Contract clauses · Spanish: 99.95% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Contract clauses · Spanish: $0.42 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Contract clauses · French: 99.38% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Contract clauses · French: $0.45 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Email triage · Czech: 97.63% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Email triage · Czech: $0.29 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Email triage · German: 97.11% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Email triage · German: $0.32 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Email triage · English (US): 98.15% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Email triage · English (US): $0.23 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Email triage · Spanish: 97.70% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Email triage · Spanish: $0.30 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Email triage · French: 97.41% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Email triage · French: $0.31 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Invoice extraction · Czech: 99.90% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Invoice extraction · Czech: $0.62 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Invoice extraction · German: 99.85% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Invoice extraction · German: $0.71 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Invoice extraction · English (US): 100.00% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Invoice extraction · English (US): $0.61 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Invoice extraction · Spanish: 99.44% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Invoice extraction · Spanish: $0.73 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Invoice extraction · French: 99.85% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Invoice extraction · French: $0.77 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Personal data detection · Czech: 99.43% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Personal data detection · Czech: $0.21 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Personal data detection · German: 99.62% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Personal data detection · German: $0.23 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Personal data detection · English (US): 99.81% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Personal data detection · English (US): $0.19 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Personal data detection · Spanish: 100.00% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Personal data detection · Spanish: $0.23 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Personal data detection · French: 99.33% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Personal data detection · French: $0.23 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Questions about a document · Czech: 99.78% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Questions about a document · Czech: $0.30 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Questions about a document · German: 100.00% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Questions about a document · German: $0.32 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Questions about a document · English (US): 100.00% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Questions about a document · English (US): $0.23 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Questions about a document · Spanish: 100.00% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Questions about a document · Spanish: $0.29 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Questions about a document · French: 99.22% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Questions about a document · French: $0.28 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Document summaries · Czech: 95.89% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Document summaries · Czech: $0.32 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Document summaries · German: 99.44% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Document summaries · German: $0.46 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Document summaries · English (US): 98.33% correct fields. Model →
- Claude Haiku 5.5 is now the cheapest model that meets the bar on Document summaries · English (US): $0.34 per 1,000 documents. Model →
- New model Claude Haiku 5.5 on Document summaries · Spanish: 94.11% correct fields. Model →
- New model Claude Haiku 5.5 on Document summaries · French: 92.78% correct fields. Model →
October 7, 2026
- New model Mistral Large 4 on Contract clauses · Czech: 100.00% correct fields. Model →
- New model Mistral Large 4 on Contract clauses · German: 99.44% correct fields. Model →
- New model Mistral Large 4 on Contract clauses · English (US): 96.10% correct fields. Model →
- New model Mistral Large 4 on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model Mistral Large 4 on Contract clauses · French: 99.28% correct fields. Model →
- New model Mistral Large 4 on Email triage · Czech: 98.15% correct fields. Model →
- New model Mistral Large 4 on Email triage · German: 97.85% correct fields. Model →
- New model Mistral Large 4 on Email triage · English (US): 98.74% correct fields. Model →
- New model Mistral Large 4 on Email triage · Spanish: 97.93% correct fields. Model →
- New model Mistral Large 4 on Email triage · French: 98.30% correct fields. Model →
- New model Mistral Large 4 on Invoice extraction · Czech: 97.08% correct fields. Model →
- New model Mistral Large 4 on Invoice extraction · German: 97.49% correct fields. Model →
- New model Mistral Large 4 on Invoice extraction · English (US): 95.39% correct fields. Model →
- New model Mistral Large 4 on Invoice extraction · Spanish: 96.51% correct fields. Model →
- New model Mistral Large 4 on Invoice extraction · French: 88.97% correct fields. Model →
- New model Mistral Large 4 on Personal data detection · Czech: 99.33% correct fields. Model →
- New model Mistral Large 4 on Personal data detection · German: 99.52% correct fields. Model →
- New model Mistral Large 4 on Personal data detection · English (US): 99.62% correct fields. Model →
- New model Mistral Large 4 on Personal data detection · Spanish: 99.81% correct fields. Model →
- New model Mistral Large 4 on Personal data detection · French: 98.67% correct fields. Model →
- New model Mistral Large 4 on Questions about a document · Czech: 89.89% correct fields. Model →
- New model Mistral Large 4 on Questions about a document · German: 93.56% correct fields. Model →
- New model Mistral Large 4 on Questions about a document · English (US): 98.89% correct fields. Model →
- New model Mistral Large 4 on Questions about a document · Spanish: 94.22% correct fields. Model →
- New model Mistral Large 4 on Questions about a document · French: 92.33% correct fields. Model →
- New model Mistral Large 4 on Document summaries · Czech: 83.33% correct fields. Model →
- New model Mistral Large 4 on Document summaries · German: 99.33% correct fields. Model →
- New model Mistral Large 4 on Document summaries · English (US): 100.00% correct fields. Model →
- New model Mistral Large 4 on Document summaries · Spanish: 91.33% correct fields. Model →
- New model Mistral Large 4 on Document summaries · French: 72.67% correct fields. Model →
September 30, 2026
- New model GLM-5.3 Flash on Contract clauses · Czech: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Contract clauses · German: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Contract clauses · English (US): 99.49% correct fields. Model →
- New model GLM-5.3 Flash on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Contract clauses · French: 99.74% correct fields. Model →
- New model GLM-5.3 Flash on Email triage · Czech: 98.89% correct fields. Model →
- New model GLM-5.3 Flash on Email triage · German: 98.00% correct fields. Model →
- New model GLM-5.3 Flash on Email triage · English (US): 99.41% correct fields. Model →
- New model GLM-5.3 Flash on Email triage · Spanish: 99.11% correct fields. Model →
- New model GLM-5.3 Flash on Email triage · French: 98.52% correct fields. Model →
- New model GLM-5.3 Flash on Invoice extraction · Czech: 99.69% correct fields. Model →
- New model GLM-5.3 Flash on Invoice extraction · German: 99.54% correct fields. Model →
- New model GLM-5.3 Flash on Invoice extraction · English (US): 99.94% correct fields. Model →
- New model GLM-5.3 Flash on Invoice extraction · Spanish: 99.38% correct fields. Model →
- New model GLM-5.3 Flash on Invoice extraction · French: 99.08% correct fields. Model →
- New model GLM-5.3 Flash on Personal data detection · Czech: 99.33% correct fields. Model →
- New model GLM-5.3 Flash on Personal data detection · German: 99.81% correct fields. Model →
- New model GLM-5.3 Flash on Personal data detection · English (US): 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Personal data detection · Spanish: 99.81% correct fields. Model →
- New model GLM-5.3 Flash on Personal data detection · French: 99.81% correct fields. Model →
- New model GLM-5.3 Flash on Questions about a document · Czech: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Questions about a document · German: 98.78% correct fields. Model →
- New model GLM-5.3 Flash on Questions about a document · English (US): 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Questions about a document · French: 99.67% correct fields. Model →
- New model GLM-5.3 Flash on Document summaries · Czech: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Document summaries · German: 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Document summaries · English (US): 100.00% correct fields. Model →
- New model GLM-5.3 Flash on Document summaries · Spanish: 96.67% correct fields. Model →
- New model GLM-5.3 Flash on Document summaries · French: 100.00% correct fields. Model →
September 29, 2026
- New model Claude Sonnet 5.5 on Contract clauses · Czech: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Contract clauses · German: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Contract clauses · English (US): 99.79% correct fields. Model →
- New model Claude Sonnet 5.5 on Contract clauses · Spanish: 100.00% correct fields. Model →
- New model Mistral Small 4 on Contract clauses · Spanish: 100.00% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Contract clauses · Spanish: $0.23 per 1,000 documents. Model →
- New model Claude Sonnet 5.5 on Contract clauses · French: 99.95% correct fields. Model →
- New model Mistral Small 4 on Contract clauses · French: 99.23% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Contract clauses · French: $0.23 per 1,000 documents. Model →
- New model Claude Sonnet 5.5 on Email triage · Czech: 98.67% correct fields. Model →
- New model Claude Sonnet 5.5 on Email triage · German: 98.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Email triage · English (US): 98.89% correct fields. Model →
- New model Claude Sonnet 5.5 on Email triage · Spanish: 97.70% correct fields. Model →
- New model Claude Sonnet 5.5 on Email triage · French: 97.78% correct fields. Model →
- New model Mistral Small 4 on Email triage · French: 90.52% correct fields. Model →
- New model Claude Sonnet 5.5 on Invoice extraction · Czech: 99.95% correct fields. Model →
- New model Claude Sonnet 5.5 on Invoice extraction · German: 99.95% correct fields. Model →
- New model Claude Sonnet 5.5 on Invoice extraction · English (US): 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Invoice extraction · Spanish: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Invoice extraction · French: 99.95% correct fields. Model →
- New model Claude Sonnet 5.5 on Personal data detection · Czech: 99.71% correct fields. Model →
- New model Claude Sonnet 5.5 on Personal data detection · German: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Personal data detection · English (US): 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Personal data detection · Spanish: 100.00% correct fields. Model →
- New model Mistral Small 4 on Personal data detection · Spanish: 99.03% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Personal data detection · Spanish: $0.09 per 1,000 documents. Model →
- New model Claude Sonnet 5.5 on Personal data detection · French: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Questions about a document · Czech: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Questions about a document · German: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Questions about a document · English (US): 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Questions about a document · Spanish: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Questions about a document · French: 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Document summaries · Czech: 98.67% correct fields. Model →
- New model Claude Sonnet 5.5 on Document summaries · German: 99.11% correct fields. Model →
- New model Claude Sonnet 5.5 on Document summaries · English (US): 100.00% correct fields. Model →
- New model Claude Sonnet 5.5 on Document summaries · Spanish: 95.56% correct fields. Model →
- New model Claude Sonnet 5.5 on Document summaries · French: 97.89% correct fields. Model →
September 28, 2026
- New model Mistral Small 4 on Questions about a document · French: 100.00% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Questions about a document · French: $0.18 per 1,000 documents. Model →
- New model Mistral Small 4 on Document summaries · Spanish: 97.78% correct fields. Model →
- New model Mistral Small 4 on Document summaries · French: 93.22% correct fields. Model →
September 28, 2026
- Claude Sonnet 5 now meets the bar on Invoice extraction · Spanish (94.62% → 98.46%). Model →
- New model Mistral Small 4 on Questions about a document · Czech: 100.00% correct fields. Model →
- New model Mistral Small 4 on Questions about a document · English (US): 100.00% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Questions about a document · English (US): $0.15 per 1,000 documents. Model →
- First results: Questions about a document · Spanish.
- First results: Questions about a document · French.
- New model Mistral Small 4 on Document summaries · German: 99.89% correct fields. Model →
- New model Mistral Small 4 on Document summaries · English (US): 99.67% correct fields. Model →
- First results: Document summaries · Spanish.
- First results: Document summaries · French.
September 28, 2026
- New model Mistral Small 4 on Document summaries · Czech: 99.07% correct fields. Model →
- New model DeepSeek V4.1 Flash on Document summaries · German: 100.00% correct fields. Model →
- New model Qwen3.8 Flash on Document summaries · German: 95.33% correct fields. Model →
- New model Qwen3.8 Flash on Document summaries · English (US): 99.89% correct fields. Model →
September 28, 2026
- First results: Questions about a document · Czech.
- First results: Questions about a document · German.
- First results: Questions about a document · English (US).
- First results: Document summaries · Czech.
- First results: Document summaries · German.
- First results: Document summaries · English (US).
September 28, 2026
- First results: Contract clauses · Spanish.
- First results: Contract clauses · French.
- First results: Email triage · Spanish.
- First results: Email triage · French.
- New model Mistral Small 4 on Invoice extraction · German: 94.82% correct fields. Model →
- First results: Invoice extraction · Spanish.
- First results: Invoice extraction · French.
- New model Mistral Small 4 on Personal data detection · German: 99.13% correct fields. Model →
- Mistral Small 4 is now the cheapest model that meets the bar on Personal data detection · German: $0.09 per 1,000 documents. Model →
- First results: Personal data detection · Spanish.
- First results: Personal data detection · French.
September 28, 2026
- Gemma 4 31B is now the cheapest model that meets the bar on Personal data detection · German: $0.10 per 1,000 documents. Model →
September 28, 2026
- First results: Contract clauses · Spanish.
- First results: Contract clauses · French.
September 28, 2026
- First results: Personal data detection · Spanish.
- First results: Personal data detection · French.
September 28, 2026
- First results: Email triage · Spanish.
- First results: Email triage · French.
September 28, 2026
- First results: Invoice extraction · Spanish.
- First results: Invoice extraction · French.
September 28, 2026
- Mistral Small 4 fell below the bar on Contract clauses · German (95.16% → 94.00%). Model →
- GPT-6 Luna is now the cheapest model that meets the bar on Contract clauses · German: $0.27 per 1,000 documents. Model →
- Gemma 4 31B now meets the bar on Email triage · German (94.81% → 95.11%). Model →
- Gemma 4 31B is now the cheapest model that meets the bar on Email triage · German: $0.15 per 1,000 documents. Model →
September 28, 2026
- First results: Contract clauses · German.
September 28, 2026
- First results: Personal data detection · German.
September 28, 2026
- First results: Email triage · German.
September 28, 2026
- First results: Contract clauses · Czech.
- First results: Contract clauses · English (US).
- First results: Invoice extraction · German.
September 28, 2026
- First results: Email triage · Czech.
- First results: Email triage · English (US).
- First results: Personal data detection · Czech.
- First results: Personal data detection · English (US).