Skip to content
Model Index
  • Overview
  • Models
  • Compare
  • Benchmarks
  • Finder
LabsMethodology
  1. Benchmarks
  2. Coding

SciCode

Independent testArtificial Analysis · updated Oct 8, 2026

What it measures

Research-grade scientific programming: 288 sub-problems drawn from real physics, chemistry, biology and maths workflows, each of which must pass all unit tests.

How to read it

Rewards getting domain science right as well as writing code. Scores move slowly — a few points is a real difference.

Open at Artificial Analysis
Best
66.9%
Claude Opus 5.5
Median
51.9%
DeepSeek V4.1 Flash

Results

38 of 41 tracked models have a result.

Filter in explorer
  1. 1
    Claude Opus 5.5

    Anthropic · Max effort

    66.9%
  2. 2
    Claude Fable 5.1

    Anthropic · Max effort

    63.1%
  3. 3
    Gemini 4 ArgonLimited

    Google DeepMind · High effort

    61.8%
  4. 4
    Claude Sonnet 5.5

    Anthropic · Max effort

    61.0%
  5. 5
    MiMo-V2.6-ProOpen

    Xiaomi

    60.9%
  6. 6
    Kimi K3Open

    Moonshot AI · Max effort

    59.5%
  7. 7
    GLM-5.3Open

    Z.ai · Max effort

    59.0%
  8. 8
    Step 5 Preview

    StepFun

    58.9%
  9. 9
    Muse Spark 1.3

    Meta · Max effort

    58.8%
  10. 10
    Gemini 3.1 Pro

    Google DeepMind

    58.7%
  11. 11
    Grok 4.7

    SpaceXAI · xHigh effort

    57.4%
  12. 12
    Gemini 3.8 Flash

    Google DeepMind · High effort

    56.6%
  13. 13
    GPT-6 Astra

    OpenAI · Max effort

    56.5%
  14. 14
    Claude Haiku 5.5

    Anthropic · Max effort

    55.0%
  15. 14
    GPT-5.6 Terra

    OpenAI · Max effort

    55.0%
  16. 16
    GPT-6 Luna

    OpenAI · Max effort

    54.6%
  17. 17
    GPT-6.1 Sol

    OpenAI · Max effort

    54.2%
  18. 17
    Mistral Large 4

    Mistral AI

    54.2%
  19. 19
    Qwen3.8-Max

    Alibaba

    52.1%
  20. 20
    DeepSeek V4.1 FlashOpen

    DeepSeek · Max effort

    51.9%
  21. 21
    GLM-5.3-FlashOpen

    Z.ai

    51.6%
  22. 22
    MiMo-V2.6-FlashOpen

    Xiaomi

    51.3%
  23. 23
    DeepSeek V4 ProOpen

    DeepSeek · Max effort

    51.0%
  24. 24
    Qwen3.8-FlashOpen

    Alibaba

    50.6%
  25. 25
    TM
    Inkling-SmallOpen

    Thinking Machines Lab

    49.7%
  26. 26
    Hy3Open

    Tencent

    48.6%
  27. 27
    Kimi K2.7 CodeOpen

    Moonshot AI

    47.8%
  28. 28
    MiniMax-M3Open

    MiniMax

    47.1%
  29. 29
    TM
    InklingOpen

    Thinking Machines Lab · xHigh effort

    47.0%
  30. 30
    Qwen3.8-27BOpen

    Alibaba · xHigh effort

    46.6%
  31. 31
    Gemma 4 31BOpen

    Google DeepMind · Reasoning effort

    45.5%
  32. 32
    Muse Glimmer 30BOpen

    Meta · High effort

    44.9%
  33. 33
    Gemini 3.5 Flash-Lite

    Google DeepMind

    41.3%
  34. 34
    Nemotron 3 UltraOpen

    NVIDIA · Reasoning effort

    40.3%
  35. 35
    Mistral Medium 3.5Open

    Mistral AI

    40.2%
  36. 36
    Mistral Small 4Open

    Mistral AI · Reasoning effort

    38.8%
  37. 37
    Command A+Open

    Cohere

    38.5%
  38. 38
    gpt-oss-120bOpen

    OpenAI · High effort

    34.0%
3 tracked models without a result

Hy4 preview, MAI-Thinking-1, Nova 2 Lite.

Missing means the evaluator has not published a run, not that the model would score poorly.

Model Index

An independent snapshot of frontier SI models and how they perform on third-party benchmarks.

Data as of Oct 8, 2026

Explore

  • All models
  • Compare
  • Benchmarks
  • Model finder
  • Labs
  • Methodology & sources

Sources

Benchmark results come from Artificial Analysis, Arena, Vals AI and ARC Prize. Specifications and prices come from each lab's own documentation.

Not affiliated with any lab or evaluator. Names and logos are trademarks of their respective owners and are used only to identify them. Scores change often — check the linked source before making a decision that depends on one.

  • Overview
  • Models
  • Compare
  • Benchmarks
  • Finder
  • Overview
  • Models
  • Compare
  • Benchmarks
  • Finder
  • Overview
  • Models
  • Compare
  • Benchmarks
  • Finder