Best AI model for contract review

One job here means a 40-page contract read against a checklist; a memo of issues written. Costs use each vendor’s list price; a product built on the model charges its own price on top. See every model and job.

1. 1

   Gemini 4 Carbon · Google

   Second on Artificial Analysis's index (60 against the leader's 65) at a quarter of the leader's list price, but only until 30 November, when Google doubles it.

   $75 per 1,000 contracts reviewedScore 75 of 100Hardest problemsIntelligence Index 60$2 read · $10 written per 1M tokensReads 1M tokens at once

   Caveat: The $2 in / $10 out per million tokens (the unit vendors bill by) is introductory; Google says $4 / $20 applies after 30 November 2026, which will lower its score here.

2. 2

   Claude Fable 5.2 · Anthropic

   The strongest model on the board for problems the others get wrong: first on every independent test read, at twice the list price of Claude Opus 5.5.

   $300 per 1,000 contracts reviewedScore 71 of 100Hardest problemsIntelligence Index 65$8 read · $40 written per 1M tokensReads 1M tokens at once

   Caveat: Anthropic's own advice is to start with Opus 5.5 and move up only when its answers fall short.

3. 3

   Claude Opus 5.5 · Anthropic

   The safer buy for hard work: six points behind Claude Fable 5.2 on Artificial Analysis's index (59 against 65) at half its list price, and the model Anthropic says to start with.

   $150 per 1,000 contracts reviewedScore 70 of 100Hardest problemsIntelligence Index 59$4 read · $20 written per 1M tokensReads 1M tokens at once

   Caveat: Twice the list price of Claude Sonnet 5.5; test your own work on Sonnet before paying for this one.

4. 4

   GPT-6.1 Sol · OpenAI

   The sensible OpenAI default: 55 on Artificial Analysis's index against 57 for GPT-6 Astra, at a fifth of Astra's list price.

   $75 per 1,000 contracts reviewedScore 70 of 100Everyday defaultIntelligence Index 55$2 read · $10 written per 1M tokensReads 1M tokens at once

   Caveat: Two points separate it from Astra on the index, which is within noise; test Astra only where Sol's answers fall short.

5. 5

   Gemini 4 Argon · Google

   People prefer its answers in blind votes more than any other model's, but it scores 50 on Artificial Analysis's index, ten behind its sibling Carbon at the same introductory price.

   $75 per 1,000 contracts reviewedScore 69 of 100Hardest problemsIntelligence Index 50$2 read · $10 written per 1M tokensReads 1M tokens at once

   Caveat: The $2 in / $10 out per million tokens (the unit vendors bill by) is introductory: Google says $4 / $20 applies from 1 December 2026, which will lower its score here.

6. 6

   Claude Sonnet 5.5 · Anthropic

   A sound everyday default: 55 on Artificial Analysis's index, four points behind Claude Opus 5.5, at half its list price.

   $75 per 1,000 contracts reviewedScore 66 of 100Everyday defaultIntelligence Index 55$2 read · $10 written per 1M tokensReads 1M tokens at once

   Caveat: The saving depends on the effort setting: at the top setting Artificial Analysis's test run cost $10.02 per task, level with Opus 5.5 ($10.04).

7. 7

   GPT-6 Astra · OpenAI

   OpenAI's most capable model and the most expensive on the board: five times the list price of GPT-6.1 Sol for two more points on Artificial Analysis's index (57 against 55).

   $375 per 1,000 contracts reviewedScore 66 of 100Hardest problemsIntelligence Index 57$10 read · $50 written per 1M tokensReads 1M tokens at once

   Caveat: The price doubles on any request that feeds it more than 272,000 tokens of text at once, such as a stack of long documents.

8. 8

   GPT-6.1 Terra · OpenAI

   Half the list price of GPT-6.1 Sol for four points less on Artificial Analysis's index (51 against 55): the cheaper OpenAI default where the work is routine.

   $38 per 1,000 contracts reviewedScore 65 of 100Everyday defaultIntelligence Index 51$1 read · $5 written per 1M tokensReads 1.05M tokens at once

   Caveat: Requests that feed it more than 272,000 tokens of text at once are billed at twice the input price and 1.5 times the output price, per OpenAI.

9. 9

   Qwen3.8 Max · Alibaba

   Alibaba's hosted flagship scores 34 on Artificial Analysis's index, a little above the open models GLM-5.3 (32) and MiMo-V2.6-Pro (30), at about 1.4 times GLM's list price and about five times MiMo's.

   $69 per 1,000 contracts reviewedScore 63 of 100Everyday defaultIntelligence Index 34$2 read · $6 written per 1M tokensReads 984K tokens at once

   Caveat: The hosted model is closed; the open version is a different model that Artificial Analysis has not scored and that needs data-centre hardware.

10. 10

    Mistral Large 4 · Mistral AI

    The European option: about a quarter of Claude Sonnet 5.5's list price and free to host yourself, but independent scores are mid-table (26 on Artificial Analysis's index).

    $24 per 1,000 contracts reviewedScore 58 of 100Everyday defaultIntelligence Index 26$0.70 read · $2.10 written per 1M tokensReads 524K tokens at once

    Caveat: Free to use and build on, but a business that resells the model as a service needs a separate commercial licence from Mistral.

11. 11

    Muse Wonder · Meta

    Meta charges Claude Opus 5.5's list price for a model that scores 44 on Artificial Analysis's index against Opus's 59: hard to justify unless images and video are the job.

    $150 per 1,000 contracts reviewedScore 58 of 100Hardest problemsIntelligence Index 44$4 read · $20 written per 1M tokensReads 1M tokens at once

    Caveat: Meta's announcement gives no price; $4 in / $20 out per million tokens is the figure Artificial Analysis lists and TechCrunch reported.

12. 12

    Grok 4.7 · xAI

    Carried by all three big clouds and listed by xAI at $2 in / $6 out per million tokens (the unit vendors bill by), but mid-table: 35 on Artificial Analysis's index against 55 for GPT-6.1 Sol.

    $69 per 1,000 contracts reviewedScore 52 of 100Everyday defaultIntelligence Index 35$2 read · $6 written per 1M tokensReads 500K tokens at once

    Caveat: It can read half as much text at once as most models on this board.

Every figure links to its source on the model profile pages. Vendor figures are labelled as vendor figures and never counted as independent tests.