Legal Benchmarks: Data Extraction leaderboard (Oct 2026): AI model scores
As of Oct 2026, the highest reported Alibaba score on Legal Benchmarks: Data Extraction is shared by Qwen 3.8 Flash (Alibaba) and Qwen 3.8 Max (Alibaba) at 56.7%, #3 on that leaderboard, per Legal Benchmarks.
What it measures
Data extraction tasks written by lawyers, each run twice; a task passes only if every criterion is met both times.
Lawyer-authored data extraction tasks with native source documents, graded with the same pass rule: every applicable criterion met in both attempts.
What the score means
Task pass rate: tasks meeting every criterion in both attempts
Share of extraction tasks that satisfy every applicable criterion in both of the model’s independent attempts.
Results from published AI leaderboards. Compare models within each benchmark; scores from different tests aren’t directly comparable. How we collect these results
Scores by model
Run by Legal Benchmarks. Leaderboard dated Oct 2026. Last checked 22 Oct 2026.
Highlight a provider: Anthropic OpenAI Google xAI Meta DeepSeek Alibaba Z.ai Moonshot MiniMax Mistral
Highlighting Alibaba models. Other providers are dimmed. Show all providers
- Anthropic
- OpenAI
- xAI
- Meta
- DeepSeek
- Alibaba
- Z.ai
- Moonshot
- MiniMax
- Mistral
- Task pass rate
- Share of extraction tasks that satisfy every applicable criterion in both of the model’s independent attempts.
- Criteria pass rate
- Share of applicable extraction criteria the model meets, as published.
- Form
- Share of applicable binary form checks (clarity, length, structure) passed in both independent attempts.
- Cost per task (USD)
- Cost per task, as published.
October 2026 round. It ranks by task pass rate, so it is not comparable with the September round (Criterion Pass²) we charted before. Provider names follow ours (Qwen is listed under Alibaba). Latency is on the source page.
All scores
| Rank | Model | Provider | Task pass rate | Criteria pass rate | Form | Cost per task (USD) |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | Anthropic | 76.7% | 91.1% | 82.2% | $2.71 |
| 2 | Claude Fable 5.1 | Anthropic | 63.3% | 89.9% | 70.6% | $1.15 |
| 3 | Claude Opus 5 | Anthropic | 56.7% | 89.2% | 67.2% | $0.75 |
| 3 | GPT-6 Astra | OpenAI | 56.7% | 88.4% | 83.5% | $0.93 |
| 3 | Qwen 3.8 Flash | Alibaba | 56.7% | 85.7% | 64.4% | $0.030 |
| 3 | Claude Haiku 5.5 | Anthropic | 56.7% | 82.3% | 77.4% | $0.20 |
| 3 | Qwen 3.8 Max | Alibaba | 56.7% | 77.6% | 62.2% | $1.51 |
| 3 | GLM-5.3 | Z.ai | 56.7% | 72.4% | 67.2% | $0.12 |
| 9 | Muse Spark 1.3 | Meta | 53.3% | 86.5% | 67.4% | $0.29 |
| 9 | GPT-6.1 Sol | OpenAI | 53.3% | 84.5% | 83.0% | $0.13 |
Show 14 more rows
| Rank | Model | Provider | Task pass rate | Criteria pass rate | Form | Cost per task (USD) |
|---|---|---|---|---|---|---|
| 9 | Grok 4.6 | xAI | 53.3% | 78.3% | 69.3% | $0.45 |
| 9 | Gemini 3.8 Flash | 53.3% | 71.7% | 66.7% | $0.21 | |
| 13 | MiniMax M3 | MiniMax | 50.0% | 79.8% | 72.8% | $0.030 |
| 13 | DeepSeek V4 Pro | DeepSeek | 50.0% | 57.1% | 67.4% | $0.15 |
| 15 | GPT-5.6 Sol | OpenAI | 46.7% | 82.0% | 78.9% | $0.20 |
| 15 | Claude Sonnet 5 | Anthropic | 46.7% | 78.8% | 70.6% | $0.26 |
| 17 | Kimi K3 | Moonshot | 43.3% | 81.8% | 75.0% | $0.49 |
| 18 | Claude Sonnet 5.5 | Anthropic | 40.0% | 83.3% | 77.6% | $0.92 |
| 18 | GPT-5.6 Terra | OpenAI | 40.0% | 78.1% | 80.9% | $0.12 |
| 18 | GPT-5.6 Luna | OpenAI | 40.0% | 77.3% | 79.6% | $0.020 |
| 18 | Mistral Large 4 | Mistral | 40.0% | 65.8% | 60.9% | $0.13 |
| 22 | DeepSeek V4 Flash | DeepSeek | 36.7% | 36.9% | 43.3% | $0.020 |
| 23 | GLM-5.3 Flash | Z.ai | 30.0% | 46.6% | 47.2% | $0.030 |
| 24 | Mistral Medium 3.5 | Mistral | 20.0% | 43.1% | 62.2% | $0.51 |
Source: Legal Benchmarks leaderboard ↗ · How the judge was calibrated (Clavia) ↗
Best-ranked Alibaba model here: Qwen 3.8 Flash (Alibaba). Browse prompts for any model →
Prompts for this kind of work
Copy-ready prompts from our library that match what Legal Benchmarks: Data Extraction tests. Run them on your own material to compare models.
AI Native Solutions @ainativesolReport or remove