SWE-bench Verified

1. 1
   Claude Opus 4.5

   as “Claude 4.5 Opus (high)”
   76.8%
2. 2
   Gemini 3 Flash

   as “Gemini 3 Flash (high)”
   75.8%
3. 2
   MiniMax-M2.5

   as “MiniMax M2.5 (high)”
   75.8%
4. 4
   Claude Opus 4.6

   as “Claude 4.6 Opus”
   75.6%
5. 5
   Gemini 3 Pro

   as “Gemini 3 Pro Preview (2025-11-18)”
   74.2%
6. 6
   GLM-5

   as “GLM 5 (high)”
   72.8%
7. 6
   GPT-5.2

   as “GPT 5.2 (high)”
   72.8%
8. 6
   GPT-5.2 Codex

   as “GPT 5.2 Codex”
   72.8%
9. 9
   Claude Sonnet 4.5

   as “Claude 4.5 Sonnet (high)”
   71.4%
10. 10
    Kimi K2.5

    as “Kimi K2.5 (high)”
    70.8%
11. 11
    DeepSeek-V3.2

    as “DeepSeek V3.2 (high)”
    70.0%
12. 12
    Claude Opus 4

    as “Claude 4 Opus (20250514)”
    67.6%
13. 13
    Claude 4.5 Haiku

    as “Claude 4.5 Haiku (high)”
    66.6%
14. 14
    GPT-5.1

    as “GPT 5.1 (2025-11-13) (medium)”
    66.0%
15. 14
    GPT 5.1 Codex

    as “GPT 5.1 Codex (medium)”
    66.0%
16. 16
    GPT-5

    as “GPT 5 (2025-08-07) (medium)”
    65.0%
17. 17
    Claude Sonnet 4

    as “Claude 4 Sonnet (20250514)”
    64.9%
18. 18
    Kimi K2 Thinking
    63.4%
19. 19
    MiniMax M2
    61.0%
20. 20
    GPT-5 mini

    as “GPT 5 mini (2025-08-07) (medium)”
    59.8%
21. 21
    o3

    as “o3 (2025-04-16)”
    58.4%
22. 22
    Devstral Small (2512)
    56.4%
23. 23
    GLM 4.6

    as “GLM 4.6 (T=1)”
    55.4%
24. 23
    Qwen3-Coder 480B/A35B Instruct
    55.4%
25. 25
    GLM 4.5

    as “GLM 4.5 (2025-08-22)”
    54.2%

History builds with every daily capture. One capture so far.

WebDev Arena

#1 Claude Opus 5.5

SWE-Bench Pro

#1 Claude Opus 5

Aider Polyglot

#1 GPT-5