Benchmark Almanac

ReactBench

Claude Opus 5.5 54.4% leads since September 2026
Not in model cards

Coding agents do 39 realistic React tasks in two tracks. Write React tasks come from merged pull requests and are checked by hidden behavior tests. Fix React tasks ask the agent to remove React performance, accessibility and quality issues. A task passes only if the tests pass and a pinned React Doctor scan finds no new graded issue.

Million also makes React Doctor, the scanner the verifier uses. The board reports pass@1 with 95% intervals, cost per rollout and results by effort level. Versions 1.0 to 1.2 are listed, and results should be compared within one version. About 281 GitHub stars. Not the same test as Qwen's in-house QwenReactBench.

Released
July 2026
Kind
Web and app front ends
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Strongest model tested
Claude Opus 5.5 #4 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
31
Entries on its own leaderboard, 28 models we track

Claude Opus 5.5 leads with 54.4%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

31 entries. Scroll sideways for the rest.

  1. 54.4 Claude Opus 5.5
  2. 49.2 Claude Fable 5.1
  3. 46.7 GPT-5.6 Sol
  4. 44.6 Claude Fable 5
  5. 42.0 Claude Opus 5
  6. 39.0 GPT-5.6 Terra
  7. 36.4 Qwen 3.8 Max
  8. 35.4 callstack/apex
  9. 34.9 DeepSeek V4.1 Flash 1017
  10. 34.4 Gemini 3.8 Flash
  11. 33.9 GPT-5.6 Luna
  12. 33.9 Grok 4.6
  13. 33.9 DeepSeek V4 Pro 0813
  14. 33.3 particle-ai/particle-1
  15. 33.3 Kimi K3
  16. 32.8 GLM-5.3-Flash
  17. 32.3 Claude Opus 4.8
  18. 32.3 Claude Sonnet 5
  19. 32.3 Grok 4.5
  20. 32.3 Grok 4.7
  21. 30.8 GLM-5.3
  22. 27.2 GLM-5.2
  23. 27.2 Muse Spark 1.1
  24. 25.6 Gemini 3.1 Pro
  25. 23.6 Kimi K2.7 Code
  26. 23.6 Muse Spark 1.2
  27. 23.1 Gemini 3.5 Flash
  28. 21.5 DeepSeek V4 Flash 0731
  29. 17.9 recursive/flint
  30. 17.4 Composer 2.5
  31. 7.69 Inkling-Small
  1. 1 Claude Opus 5.5 v1.2; max 54.4%
  2. 2 Claude Fable 5.1 v1.2; max 49.2%
  3. 3 GPT-5.6 Sol v1.2; max 46.7%
  4. 4 Claude Fable 5 v1.2; max 44.6%
  5. 5 Claude Opus 5 v1.2; max 42%
  6. 6 GPT-5.6 Terra v1.2; max 39%
  7. 7 Qwen 3.8 Max v1.2; xhigh 36.4%
  8. 8 callstack/apex v1.2; medium 35.4%
  9. 9 DeepSeek V4.1 Flash 1017 v1.2; medium 34.9%
  10. 10 Gemini 3.8 Flash v1.2; max 34.4%
  11. 11 GPT-5.6 Luna v1.2; max 33.9%
  12. 12 Grok 4.6 v1.2; high 33.9%
  13. 13 DeepSeek V4 Pro 0813 v1.2; max 33.9%
  14. 14 particle-ai/particle-1 v1.2; medium 33.3%
  15. 15 Kimi K3 v1.2; mini-swe-agent-preinstalled 33.3%
  16. 16 GLM-5.3-Flash v1.2; max 32.8%
  17. 17 Claude Opus 4.8 v1.2; max 32.3%
  18. 18 Claude Sonnet 5 v1.2; xhigh 32.3%
  19. 19 Grok 4.5 v1.2; low 32.3%
  20. 20 Grok 4.7 v1.2; low 32.3%
  21. 21 GLM-5.3 v1.2; max 30.8%
  22. 22 GLM-5.2 v1.2; max 27.2%
  23. 23 Muse Spark 1.1 v1.2; max 27.2%
  24. 24 Gemini 3.1 Pro v1.2; high 25.6%
  25. 25 Kimi K2.7 Code v1.2; mini-swe-agent-preinstalled 23.6%
  26. 26 Muse Spark 1.2 v1.2; xhigh 23.6%
  27. 27 Gemini 3.5 Flash v1.2; medium 23.1%
  28. 28 DeepSeek V4 Flash 0731 v1.2; max 21.5%
  29. 29 recursive/flint v1.2; mini-swe-agent-preinstalled 17.9%
  30. 30 Composer 2.5 v1.2; cursor-cli 17.4%
  31. 31 Inkling-Small v1.2; thinking-0.99 7.69%

Source: Official ReactBench leaderboard

#EntryScoreSettingDateSource
1 Claude Opus 5.5 54.4% v1.2; max Sep 2026 Official leaderboard
2 Claude Fable 5.1 49.2% v1.2; max Sep 2026 Official leaderboard
3 GPT-5.6 Sol 46.7% v1.2; max Jul 2026 Official leaderboard
4 Claude Fable 5 44.6% v1.2; max Jun 2026 Official leaderboard
5 Claude Opus 5 42% v1.2; max Jul 2026 Official leaderboard
6 GPT-5.6 Terra 39% v1.2; max Jul 2026 Official leaderboard
7 Qwen 3.8 Max 36.4% v1.2; xhigh Aug 2026 Official leaderboard
8 callstack/apex 35.4% v1.2; medium Official leaderboard
9 DeepSeek V4.1 Flash 1017 34.9% v1.2; medium Oct 2026 Official leaderboard
10 Gemini 3.8 Flash 34.4% v1.2; max Sep 2026 Official leaderboard
11 GPT-5.6 Luna 33.9% v1.2; max Jul 2026 Official leaderboard
12 Grok 4.6 33.9% v1.2; high, Cursor Aug 2026 Official leaderboard
13 DeepSeek V4 Pro 0813 33.9% v1.2; max Aug 2026 Official leaderboard
14 particle-ai/particle-1 33.3% v1.2; medium Official leaderboard
15 Kimi K3 33.3% v1.2; mini-swe-agent-preinstalled, mini-swe-agent Jul 2026 Official leaderboard
16 GLM-5.3-Flash 32.8% v1.2; max Aug 2026 Official leaderboard
17 Claude Opus 4.8 32.3% v1.2; max May 2026 Official leaderboard
18 Claude Sonnet 5 32.3% v1.2; xhigh Jun 2026 Official leaderboard
19 Grok 4.5 32.3% v1.2; low, Cursor Jul 2026 Official leaderboard
20 Grok 4.7 32.3% v1.2; low, Cursor Sep 2026 Official leaderboard
21 GLM-5.3 30.8% v1.2; max Aug 2026 Official leaderboard
22 GLM-5.2 27.2% v1.2; max Jun 2026 Official leaderboard
23 Muse Spark 1.1 27.2% v1.2; max Jul 2026 Official leaderboard
24 Gemini 3.1 Pro 25.6% v1.2; high Feb 2026 Official leaderboard
25 Kimi K2.7 Code 23.6% v1.2; mini-swe-agent-preinstalled, mini-swe-agent Jun 2026 Official leaderboard
26 Muse Spark 1.2 23.6% v1.2; xhigh Aug 2026 Official leaderboard
27 Gemini 3.5 Flash 23.1% v1.2; medium May 2026 Official leaderboard
28 DeepSeek V4 Flash 0731 21.5% v1.2; max Jul 2026 Official leaderboard
29 recursive/flint 17.9% v1.2; mini-swe-agent-preinstalled, mini-swe-agent Official leaderboard
30 Composer 2.5 17.4% v1.2; cursor-cli May 2026 Official leaderboard
31 Inkling-Small 7.69% v1.2; thinking-0.99 Jul 2026 Official leaderboard

ReactBench score vs. cost

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 20% 40% 60% $0$5$10$15 Cost per task ($), lower is better Pass rate Gemini 3.1 Pro (v1.2; high): Pass rate 25.6%, cost per task $0.8 Gemini 3.1 Pro (v1.2; xhigh): Pass rate 23.1%, cost per task $0.8 Gemini 3.1 Pro (v1.2; max): Pass rate 22.1%, cost per task $0.84 Gemini 3.1 Pro (v1.2; medium): Pass rate 21.5%, cost per task $0.84 Gemini 3.1 Pro (v1.2; low): Pass rate 12.8%, cost per task $0.37 Composer 2.5 (v1.2; cursor-cli): Pass rate 17.4%, cost per task $0.14 Gemini 3.5 Flash (v1.2; medium): Pass rate 23.1%, cost per task $1.4 Gemini 3.5 Flash (v1.2; xhigh): Pass rate 22.6%, cost per task $1.44 Gemini 3.5 Flash (v1.2; high): Pass rate 21.5%, cost per task $1.43 Gemini 3.5 Flash (v1.2; low): Pass rate 21.5%, cost per task $0.92 Claude Opus 4.8 (v1.2; max): Pass rate 32.3%, cost per task $6.77 Claude Opus 4.8 (v1.2; xhigh): Pass rate 30.8%, cost per task $5.63 Claude Opus 4.8 (v1.2; medium): Pass rate 28.2%, cost per task $3.41 Claude Opus 4.8 (v1.2; high): Pass rate 27.7%, cost per task $4.08 Claude Opus 4.8 (v1.2; low): Pass rate 22.1%, cost per task $1.84 Claude Fable 5 (v1.2; max): Pass rate 44.6%, cost per task $12.8 Claude Fable 5 (v1.2; xhigh): Pass rate 43.1%, cost per task $9.17 Claude Fable 5 (v1.2; high): Pass rate 40.5%, cost per task $6.8 Claude Fable 5 (v1.2; low): Pass rate 35.4%, cost per task $3.05 Kimi K2.7 Code (v1.2; mini-swe-agent-preinstalled): Pass rate 23.6%, cost per task $1.27 GLM-5.2 (v1.2; max): Pass rate 27.2%, cost per task $1.23 Claude Sonnet 5 (v1.2; xhigh): Pass rate 32.3%, cost per task $3.17 Claude Sonnet 5 (v1.2; max): Pass rate 31.3%, cost per task $4.74 Claude Sonnet 5 (v1.2; high): Pass rate 30.3%, cost per task $2.31 Claude Sonnet 5 (v1.2; medium): Pass rate 22.6%, cost per task $1.44 Claude Sonnet 5 (v1.2; low): Pass rate 16.9%, cost per task $0.74 Grok 4.5 (v1.2; low): Pass rate 32.3%, cost per task $0.95 Grok 4.5 (v1.2; high): Pass rate 31.8%, cost per task $0.95 Grok 4.5 (v1.2; medium): Pass rate 30.8%, cost per task $0.96 GPT-5.6 Luna (v1.2; max): Pass rate 33.9%, cost per task $0.84 GPT-5.6 Luna (v1.2; high): Pass rate 32.3%, cost per task $0.38 GPT-5.6 Luna (v1.2; xhigh): Pass rate 31.3%, cost per task $0.57 GPT-5.6 Luna (v1.2; medium): Pass rate 20%, cost per task $0.15 GPT-5.6 Luna (v1.2; low): Pass rate 10.8%, cost per task $0.07 GPT-5.6 Sol (v1.2; max): Pass rate 46.7%, cost per task $3.06 GPT-5.6 Sol (v1.2; xhigh): Pass rate 41%, cost per task $2.02 GPT-5.6 Sol (v1.2; high): Pass rate 36.9%, cost per task $1.58 GPT-5.6 Sol (v1.2; medium): Pass rate 34.9%, cost per task $1.1 GPT-5.6 Sol (v1.2; low): Pass rate 33.9%, cost per task $0.59 GPT-5.6 Terra (v1.2; max): Pass rate 39%, cost per task $1.13 GPT-5.6 Terra (v1.2; xhigh): Pass rate 36.9%, cost per task $0.6 GPT-5.6 Terra (v1.2; medium): Pass rate 29.7%, cost per task $0.25 GPT-5.6 Terra (v1.2; low): Pass rate 28.2%, cost per task $0.21 Muse Spark 1.1 (v1.2; max): Pass rate 27.2%, cost per task $2.69 Muse Spark 1.1 (v1.2; high): Pass rate 25.6%, cost per task $2.21 Muse Spark 1.1 (v1.2; medium): Pass rate 25.1%, cost per task $1.3 Muse Spark 1.1 (v1.2; xhigh): Pass rate 25.1%, cost per task $2.57 Muse Spark 1.1 (v1.2; low): Pass rate 17.9%, cost per task $0.65 Inkling-Small (v1.2; thinking-0.99): Pass rate 7.69%, cost per task $0.1 Kimi K3 (v1.2; mini-swe-agent-preinstalled): Pass rate 33.3%, cost per task $1.92 Claude Opus 5 (v1.2; max): Pass rate 42%, cost per task $5.94 Claude Opus 5 (v1.2; high): Pass rate 41.5%, cost per task $3.68 Claude Opus 5 (v1.2; xhigh): Pass rate 36.4%, cost per task $4.93 Claude Opus 5 (v1.2; medium): Pass rate 34.4%, cost per task $2.04 Claude Opus 5 (v1.2; low): Pass rate 27.2%, cost per task $1.19 DeepSeek V4 Flash 0731 (v1.2; max): Pass rate 21.5%, cost per task $0.26 Qwen 3.8 Max (v1.2; xhigh): Pass rate 36.4%, cost per task $1.62 Muse Spark 1.2 (v1.2; xhigh): Pass rate 23.6%, cost per task $4.67 Grok 4.6 (v1.2; high): Pass rate 33.9%, cost per task $1.68 Grok 4.6 (v1.2; xhigh): Pass rate 33.9%, cost per task $2.04 Grok 4.6 (v1.2; medium): Pass rate 29.2%, cost per task $1.02 Grok 4.6 (v1.2; low): Pass rate 28.7%, cost per task $0.66 DeepSeek V4 Pro 0813 (v1.2; max): Pass rate 33.9%, cost per task $0.09 GLM-5.3 (v1.2; max): Pass rate 30.8%, cost per task $2.25 Claude Fable 5.1 (v1.2; max): Pass rate 49.2%, cost per task $5.5 Claude Fable 5.1 (v1.2; xhigh): Pass rate 46.1%, cost per task $3.74 Claude Fable 5.1 (v1.2; high): Pass rate 44.1%, cost per task $1.74 Claude Fable 5.1 (v1.2; medium): Pass rate 33.9%, cost per task $1.12 Claude Fable 5.1 (v1.2; low): Pass rate 32.8%, cost per task $0.84 Gemini 3.8 Flash (v1.2; max): Pass rate 34.4%, cost per task $1.63 Grok 4.7 (v1.2; low): Pass rate 32.3%, cost per task $0.68 Grok 4.7 (v1.2; medium): Pass rate 31.8%, cost per task $2.42 Grok 4.7 (v1.2; xhigh): Pass rate 31.8%, cost per task $4.56 Grok 4.7 (v1.2; high): Pass rate 27.7%, cost per task $3.46 Claude Opus 5.5 (v1.2; low): Pass rate 30.8%, cost per task $0.24 DeepSeek V4.1 Flash 1017 (v1.2; high): Pass rate 32.3%, cost per task $0.09 DeepSeek V4.1 Flash 1017 (v1.2; minimal): Pass rate 30.3%, cost per task $0.07 DeepSeek V4.1 Flash 1017 (v1.2; max): Pass rate 29.2%, cost per task $0.09 callstack/apex (v1.2; medium): Pass rate 35.4%, cost per task $1.42. Not in our model list callstack/apex (v1.2; xhigh): Pass rate 34.4%, cost per task $1.98. Not in our model list particle-ai/particle-1 (v1.2; medium): Pass rate 33.3%, cost per task $0.18. Not in our model list particle-ai/particle-1 (v1.2; high): Pass rate 32.8%, cost per task $0.17. Not in our model list callstack/apex (v1.2; low): Pass rate 29.7%, cost per task $1.54. Not in our model list particle-ai/particle-1 (v1.2; low): Pass rate 29.2%, cost per task $0.12. Not in our model list particle-ai/particle-1 (v1.2; max): Pass rate 29.2%, cost per task $0.18. Not in our model list particle-ai/particle-1 (v1.2; xhigh): Pass rate 28.2%, cost per task $0.17. Not in our model list GPT-5.6 Terra (v1.2; high): Pass rate 37.4%, cost per task $0.4 GLM-5.3-Flash (v1.2; max): Pass rate 32.8%, cost per task $0.07 Claude Opus 5.5 (v1.2; max): Pass rate 54.4%, cost per task $5.64 Claude Opus 5.5 (v1.2; xhigh): Pass rate 50.8%, cost per task $1.62 Claude Opus 5.5 (v1.2; high): Pass rate 46.1%, cost per task $0.67 Claude Opus 5.5 (v1.2; medium): Pass rate 43.6%, cost per task $0.49 DeepSeek V4.1 Flash 1017 (v1.2; medium): Pass rate 34.9%, cost per task $0.1 DeepSeek V4.1 Flash 1017 (v1.2; low): Pass rate 34.4%, cost per task $0.07 Claude Opus 5.5 (v1.2; max)Claude Opus 5.5 (v1.2; xhigh)Claude Opus 5.5 (v1.2; high)Claude Opus 5.5 (v1.2; medium)GPT-5.6 Terra (v1.2; high)DeepSeek V4.1 Flash 1017 (v1.2; medium)DeepSeek V4.1 Flash 1017 (v1.2; low)GLM-5.3-Flash (v1.2; max) 0% 20% 40% 60% $0$5$10$15 Cost per task ($), lower is better Pass rate Gemini 3.1 Pro (v1.2; high): Pass rate 25.6%, cost per task $0.8 Gemini 3.1 Pro (v1.2; xhigh): Pass rate 23.1%, cost per task $0.8 Gemini 3.1 Pro (v1.2; max): Pass rate 22.1%, cost per task $0.84 Gemini 3.1 Pro (v1.2; medium): Pass rate 21.5%, cost per task $0.84 Gemini 3.1 Pro (v1.2; low): Pass rate 12.8%, cost per task $0.37 Composer 2.5 (v1.2; cursor-cli): Pass rate 17.4%, cost per task $0.14 Gemini 3.5 Flash (v1.2; medium): Pass rate 23.1%, cost per task $1.4 Gemini 3.5 Flash (v1.2; xhigh): Pass rate 22.6%, cost per task $1.44 Gemini 3.5 Flash (v1.2; high): Pass rate 21.5%, cost per task $1.43 Gemini 3.5 Flash (v1.2; low): Pass rate 21.5%, cost per task $0.92 Claude Opus 4.8 (v1.2; max): Pass rate 32.3%, cost per task $6.77 Claude Opus 4.8 (v1.2; xhigh): Pass rate 30.8%, cost per task $5.63 Claude Opus 4.8 (v1.2; medium): Pass rate 28.2%, cost per task $3.41 Claude Opus 4.8 (v1.2; high): Pass rate 27.7%, cost per task $4.08 Claude Opus 4.8 (v1.2; low): Pass rate 22.1%, cost per task $1.84 Claude Fable 5 (v1.2; max): Pass rate 44.6%, cost per task $12.8 Claude Fable 5 (v1.2; xhigh): Pass rate 43.1%, cost per task $9.17 Claude Fable 5 (v1.2; high): Pass rate 40.5%, cost per task $6.8 Claude Fable 5 (v1.2; low): Pass rate 35.4%, cost per task $3.05 Kimi K2.7 Code (v1.2; mini-swe-agent-preinstalled): Pass rate 23.6%, cost per task $1.27 GLM-5.2 (v1.2; max): Pass rate 27.2%, cost per task $1.23 Claude Sonnet 5 (v1.2; xhigh): Pass rate 32.3%, cost per task $3.17 Claude Sonnet 5 (v1.2; max): Pass rate 31.3%, cost per task $4.74 Claude Sonnet 5 (v1.2; high): Pass rate 30.3%, cost per task $2.31 Claude Sonnet 5 (v1.2; medium): Pass rate 22.6%, cost per task $1.44 Claude Sonnet 5 (v1.2; low): Pass rate 16.9%, cost per task $0.74 Grok 4.5 (v1.2; low): Pass rate 32.3%, cost per task $0.95 Grok 4.5 (v1.2; high): Pass rate 31.8%, cost per task $0.95 Grok 4.5 (v1.2; medium): Pass rate 30.8%, cost per task $0.96 GPT-5.6 Luna (v1.2; max): Pass rate 33.9%, cost per task $0.84 GPT-5.6 Luna (v1.2; high): Pass rate 32.3%, cost per task $0.38 GPT-5.6 Luna (v1.2; xhigh): Pass rate 31.3%, cost per task $0.57 GPT-5.6 Luna (v1.2; medium): Pass rate 20%, cost per task $0.15 GPT-5.6 Luna (v1.2; low): Pass rate 10.8%, cost per task $0.07 GPT-5.6 Sol (v1.2; max): Pass rate 46.7%, cost per task $3.06 GPT-5.6 Sol (v1.2; xhigh): Pass rate 41%, cost per task $2.02 GPT-5.6 Sol (v1.2; high): Pass rate 36.9%, cost per task $1.58 GPT-5.6 Sol (v1.2; medium): Pass rate 34.9%, cost per task $1.1 GPT-5.6 Sol (v1.2; low): Pass rate 33.9%, cost per task $0.59 GPT-5.6 Terra (v1.2; max): Pass rate 39%, cost per task $1.13 GPT-5.6 Terra (v1.2; xhigh): Pass rate 36.9%, cost per task $0.6 GPT-5.6 Terra (v1.2; medium): Pass rate 29.7%, cost per task $0.25 GPT-5.6 Terra (v1.2; low): Pass rate 28.2%, cost per task $0.21 Muse Spark 1.1 (v1.2; max): Pass rate 27.2%, cost per task $2.69 Muse Spark 1.1 (v1.2; high): Pass rate 25.6%, cost per task $2.21 Muse Spark 1.1 (v1.2; medium): Pass rate 25.1%, cost per task $1.3 Muse Spark 1.1 (v1.2; xhigh): Pass rate 25.1%, cost per task $2.57 Muse Spark 1.1 (v1.2; low): Pass rate 17.9%, cost per task $0.65 Inkling-Small (v1.2; thinking-0.99): Pass rate 7.69%, cost per task $0.1 Kimi K3 (v1.2; mini-swe-agent-preinstalled): Pass rate 33.3%, cost per task $1.92 Claude Opus 5 (v1.2; max): Pass rate 42%, cost per task $5.94 Claude Opus 5 (v1.2; high): Pass rate 41.5%, cost per task $3.68 Claude Opus 5 (v1.2; xhigh): Pass rate 36.4%, cost per task $4.93 Claude Opus 5 (v1.2; medium): Pass rate 34.4%, cost per task $2.04 Claude Opus 5 (v1.2; low): Pass rate 27.2%, cost per task $1.19 DeepSeek V4 Flash 0731 (v1.2; max): Pass rate 21.5%, cost per task $0.26 Qwen 3.8 Max (v1.2; xhigh): Pass rate 36.4%, cost per task $1.62 Muse Spark 1.2 (v1.2; xhigh): Pass rate 23.6%, cost per task $4.67 Grok 4.6 (v1.2; high): Pass rate 33.9%, cost per task $1.68 Grok 4.6 (v1.2; xhigh): Pass rate 33.9%, cost per task $2.04 Grok 4.6 (v1.2; medium): Pass rate 29.2%, cost per task $1.02 Grok 4.6 (v1.2; low): Pass rate 28.7%, cost per task $0.66 DeepSeek V4 Pro 0813 (v1.2; max): Pass rate 33.9%, cost per task $0.09 GLM-5.3 (v1.2; max): Pass rate 30.8%, cost per task $2.25 Claude Fable 5.1 (v1.2; max): Pass rate 49.2%, cost per task $5.5 Claude Fable 5.1 (v1.2; xhigh): Pass rate 46.1%, cost per task $3.74 Claude Fable 5.1 (v1.2; high): Pass rate 44.1%, cost per task $1.74 Claude Fable 5.1 (v1.2; medium): Pass rate 33.9%, cost per task $1.12 Claude Fable 5.1 (v1.2; low): Pass rate 32.8%, cost per task $0.84 Gemini 3.8 Flash (v1.2; max): Pass rate 34.4%, cost per task $1.63 Grok 4.7 (v1.2; low): Pass rate 32.3%, cost per task $0.68 Grok 4.7 (v1.2; medium): Pass rate 31.8%, cost per task $2.42 Grok 4.7 (v1.2; xhigh): Pass rate 31.8%, cost per task $4.56 Grok 4.7 (v1.2; high): Pass rate 27.7%, cost per task $3.46 Claude Opus 5.5 (v1.2; low): Pass rate 30.8%, cost per task $0.24 DeepSeek V4.1 Flash 1017 (v1.2; high): Pass rate 32.3%, cost per task $0.09 DeepSeek V4.1 Flash 1017 (v1.2; minimal): Pass rate 30.3%, cost per task $0.07 DeepSeek V4.1 Flash 1017 (v1.2; max): Pass rate 29.2%, cost per task $0.09 callstack/apex (v1.2; medium): Pass rate 35.4%, cost per task $1.42. Not in our model list callstack/apex (v1.2; xhigh): Pass rate 34.4%, cost per task $1.98. Not in our model list particle-ai/particle-1 (v1.2; medium): Pass rate 33.3%, cost per task $0.18. Not in our model list particle-ai/particle-1 (v1.2; high): Pass rate 32.8%, cost per task $0.17. Not in our model list callstack/apex (v1.2; low): Pass rate 29.7%, cost per task $1.54. Not in our model list particle-ai/particle-1 (v1.2; low): Pass rate 29.2%, cost per task $0.12. Not in our model list particle-ai/particle-1 (v1.2; max): Pass rate 29.2%, cost per task $0.18. Not in our model list particle-ai/particle-1 (v1.2; xhigh): Pass rate 28.2%, cost per task $0.17. Not in our model list GPT-5.6 Terra (v1.2; high): Pass rate 37.4%, cost per task $0.4 GLM-5.3-Flash (v1.2; max): Pass rate 32.8%, cost per task $0.07 Claude Opus 5.5 (v1.2; max): Pass rate 54.4%, cost per task $5.64 Claude Opus 5.5 (v1.2; xhigh): Pass rate 50.8%, cost per task $1.62 Claude Opus 5.5 (v1.2; high): Pass rate 46.1%, cost per task $0.67 Claude Opus 5.5 (v1.2; medium): Pass rate 43.6%, cost per task $0.49 DeepSeek V4.1 Flash 1017 (v1.2; medium): Pass rate 34.9%, cost per task $0.1 DeepSeek V4.1 Flash 1017 (v1.2; low): Pass rate 34.4%, cost per task $0.07 Claude Opus 5.5 (v1.2; max)Claude Opus 5.5 (v1.2; xhigh)Claude Opus 5.5 (v1.2; high)Claude Opus 5.5 (v1.2; medium)GPT-5.6 Terra (v1.2; high)DeepSeek V4.1 Flash 1017 (v1.2; medium)DeepSeek V4.1 Flash 1017 (v1.2; low)GLM-5.3-Flash (v1.2; max)
ModelSettingPass rateCost per task
Claude Opus 5.5 v1.2; max 54.4% $5.64
Claude Opus 5.5 v1.2; xhigh 50.8% $1.62
Claude Fable 5.1 v1.2; max 49.2% $5.5
GPT-5.6 Sol v1.2; max 46.7% $3.06
Claude Opus 5.5 v1.2; high 46.2% $0.67
Claude Fable 5.1 v1.2; xhigh 46.2% $3.74
Claude Fable 5 v1.2; max 44.6% $12.76
Claude Fable 5.1 v1.2; high 44.1% $1.74
Claude Opus 5.5 v1.2; medium 43.6% $0.49
Claude Fable 5 v1.2; xhigh 43.1% $9.17
Claude Opus 5 v1.2; max 42.1% $5.94
Claude Opus 5 v1.2; high 41.5% $3.68
GPT-5.6 Sol v1.2; xhigh 41% $2.02
Claude Fable 5 v1.2; high 40.5% $6.8
GPT-5.6 Terra v1.2; max 39% $1.13
GPT-5.6 Terra v1.2; high 37.4% $0.4
GPT-5.6 Terra v1.2; xhigh 36.9% $0.6
GPT-5.6 Sol v1.2; high 36.9% $1.58
Qwen 3.8 Max v1.2; xhigh 36.4% $1.62
Claude Opus 5 v1.2; xhigh 36.4% $4.93
callstack/apex v1.2; medium 35.4% $1.42
Claude Fable 5 v1.2; low 35.4% $3.05
DeepSeek V4.1 Flash 1017 v1.2; medium 34.9% $0.1
GPT-5.6 Sol v1.2; medium 34.9% $1.1
DeepSeek V4.1 Flash 1017 v1.2; low 34.4% $0.07
Gemini 3.8 Flash v1.2; max 34.4% $1.63
callstack/apex v1.2; xhigh 34.4% $1.98
Claude Opus 5 v1.2; medium 34.4% $2.04
DeepSeek V4 Pro 0813 v1.2; max 33.9% $0.09
GPT-5.6 Sol v1.2; low 33.9% $0.59
GPT-5.6 Luna v1.2; max 33.9% $0.84
Claude Fable 5.1 v1.2; medium 33.9% $1.12
Grok 4.6 v1.2; high, Cursor 33.9% $1.68
Grok 4.6 v1.2; xhigh, Cursor 33.9% $2.04
particle-ai/particle-1 v1.2; medium 33.3% $0.18
Kimi K3 v1.2; mini-swe-agent-preinstalled, mini-swe-agent 33.3% $1.92
GLM-5.3-Flash v1.2; max 32.8% $0.07
particle-ai/particle-1 v1.2; high 32.8% $0.18
Claude Fable 5.1 v1.2; low 32.8% $0.84
DeepSeek V4.1 Flash 1017 v1.2; high 32.3% $0.09
GPT-5.6 Luna v1.2; high 32.3% $0.38
Grok 4.7 v1.2; low, Cursor 32.3% $0.68
Grok 4.5 v1.2; low, Cursor 32.3% $0.95
Claude Sonnet 5 v1.2; xhigh 32.3% $3.17
Claude Opus 4.8 v1.2; max 32.3% $6.77
Grok 4.5 v1.2; high, Cursor 31.8% $0.95
Grok 4.7 v1.2; medium, Cursor 31.8% $2.42
Grok 4.7 v1.2; xhigh, Cursor 31.8% $4.56
GPT-5.6 Luna v1.2; xhigh 31.3% $0.57
Claude Sonnet 5 v1.2; max 31.3% $4.74
Claude Opus 5.5 v1.2; low 30.8% $0.24
Grok 4.5 v1.2; medium, Cursor 30.8% $0.96
GLM-5.3 v1.2; max 30.8% $2.25
Claude Opus 4.8 v1.2; xhigh 30.8% $5.63
DeepSeek V4.1 Flash 1017 v1.2; minimal 30.3% $0.07
Claude Sonnet 5 v1.2; high 30.3% $2.31
GPT-5.6 Terra v1.2; medium 29.7% $0.25
callstack/apex v1.2; low 29.7% $1.54
DeepSeek V4.1 Flash 1017 v1.2; max 29.2% $0.09
particle-ai/particle-1 v1.2; low 29.2% $0.12
particle-ai/particle-1 v1.2; max 29.2% $0.18
Grok 4.6 v1.2; medium, Cursor 29.2% $1.02
Grok 4.6 v1.2; low, Cursor 28.7% $0.66
particle-ai/particle-1 v1.2; xhigh 28.2% $0.17
GPT-5.6 Terra v1.2; low 28.2% $0.21
Claude Opus 4.8 v1.2; medium 28.2% $3.41
Grok 4.7 v1.2; high, Cursor 27.7% $3.46
Claude Opus 4.8 v1.2; high 27.7% $4.08
Claude Opus 5 v1.2; low 27.2% $1.19
GLM-5.2 v1.2; max 27.2% $1.23
Muse Spark 1.1 v1.2; max 27.2% $2.69
Gemini 3.1 Pro v1.2; high 25.6% $0.8
Muse Spark 1.1 v1.2; high 25.6% $2.21
Muse Spark 1.1 v1.2; medium 25.1% $1.3
Muse Spark 1.1 v1.2; xhigh 25.1% $2.57
Kimi K2.7 Code v1.2; mini-swe-agent-preinstalled, mini-swe-agent 23.6% $1.27
Muse Spark 1.2 v1.2; xhigh 23.6% $4.67
Gemini 3.1 Pro v1.2; xhigh 23.1% $0.8
Gemini 3.5 Flash v1.2; medium 23.1% $1.4
Claude Sonnet 5 v1.2; medium 22.6% $1.44
Gemini 3.5 Flash v1.2; xhigh 22.6% $1.44
Gemini 3.1 Pro v1.2; max 22.1% $0.84
Claude Opus 4.8 v1.2; low 22.1% $1.85
DeepSeek V4 Flash 0731 v1.2; max 21.5% $0.26
Gemini 3.1 Pro v1.2; medium 21.5% $0.84
Gemini 3.5 Flash v1.2; low 21.5% $0.92
Gemini 3.5 Flash v1.2; high 21.5% $1.43
GPT-5.6 Luna v1.2; medium 20% $0.15
Muse Spark 1.1 v1.2; low 18% $0.65
Composer 2.5 v1.2; cursor-cli 17.4% $0.14
Claude Sonnet 5 v1.2; low 16.9% $0.74
Gemini 3.1 Pro v1.2; low 12.8% $0.37
GPT-5.6 Luna v1.2; low 10.8% $0.07
Inkling-Small v1.2; thinking-0.99 7.7% $0.1

The best score, 54.4%, costs $5.64 a task

Runs on the cost frontier score higher than any run that's cheaper. Claude Opus 5.5 at v1.2; xhigh comes within 3.6 points of the best for $1.62 a task, 3 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Z.ai Anthropic DeepSeek Cost frontier One model at each effort level
0% 20% 40% 60% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 5.5, v1.2; max Claude Opus 5.5, v1.2; xhigh Claude Opus 5.5, v1.2; high Claude Opus 5.5, v1.2; medium GPT-5.6 Terra, v1.2; high DeepSeek V4.1 Flash 1017, v1.2; medium GLM-5.3-Flash, v1.2; max 0% 20% 40% 60% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 5.5, v1.2; max Claude Opus 5.5, v1.2; xhigh

Source: Official ReactBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 5.5 v1.2; max 54.4% $5.64 131,659
Claude Opus 5.5 v1.2; xhigh 50.8% $1.62 39,634
Claude Fable 5.1 v1.2; max 49.2% $5.50 69,979
GPT-5.6 Sol v1.2; max 46.7% $3.06 26,024
Claude Fable 5.1 v1.2; xhigh 46.1% $3.74 47,169
Claude Opus 5.5 v1.2; high 46.1% $0.67 15,216
Claude Fable 5 v1.2; max 44.6% $13 57,657
Claude Fable 5.1 v1.2; high 44.1% $1.74 21,895
Claude Opus 5.5 v1.2; medium 43.6% $0.49 10,675
Claude Fable 5 v1.2; xhigh 43.1% $9.17 39,290
Claude Opus 5 v1.2; max 42% $5.94 77,951
Claude Opus 5 v1.2; high 41.5% $3.68 43,679
GPT-5.6 Sol v1.2; xhigh 41% $2.02 17,532
Claude Fable 5 v1.2; high 40.5% $6.80 28,241
GPT-5.6 Terra v1.2; max 39% $1.13 29,926
GPT-5.6 Terra v1.2; high 37.4% $0.40 10,386
GPT-5.6 Sol v1.2; high 36.9% $1.58 13,452
GPT-5.6 Terra v1.2; xhigh 36.9% $0.60 16,463
Claude Opus 5 v1.2; xhigh 36.4% $4.93 59,938
Qwen 3.8 Max v1.2; xhigh 36.4% $1.62 55,024
Claude Fable 5 v1.2; low 35.4% $3.05 12,399
GPT-5.6 Sol v1.2; medium 34.9% $1.10 8,830
DeepSeek V4.1 Flash 1017 v1.2; medium 34.9% $0.10 36,115
Claude Opus 5 v1.2; medium 34.4% $2.04 22,890
Gemini 3.8 Flash v1.2; max 34.4% $1.63 100,156
DeepSeek V4.1 Flash 1017 v1.2; low 34.4% $0.07 28,041
GPT-5.6 Luna v1.2; max 33.9% $0.84 31,601
GPT-5.6 Sol v1.2; low 33.9% $0.59 4,681
Grok 4.6 v1.2; high Cursor 33.9% $1.68 37,800
Grok 4.6 v1.2; xhigh Cursor 33.9% $2.04 45,738
DeepSeek V4 Pro 0813 v1.2; max 33.9% $0.09 45,854
Claude Fable 5.1 v1.2; medium 33.9% $1.12 13,670
Kimi K3 v1.2; mini-swe-agent-preinstalled mini-swe-agent 33.3% $1.92 38,119
GLM-5.3-Flash v1.2; max 32.8% $0.07 244
Claude Fable 5.1 v1.2; low 32.8% $0.84 10,292
Claude Opus 4.8 v1.2; max 32.3% $6.77 72,900
Claude Sonnet 5 v1.2; xhigh 32.3% $3.17 47,752
Grok 4.5 v1.2; low Cursor 32.3% $0.95 20,619
GPT-5.6 Luna v1.2; high 32.3% $0.38 15,721
Grok 4.7 v1.2; low Cursor 32.3% $0.68 16,552
DeepSeek V4.1 Flash 1017 v1.2; high 32.3% $0.09 35,266
Grok 4.5 v1.2; high Cursor 31.8% $0.95 20,921
Grok 4.7 v1.2; medium Cursor 31.8% $2.42 56,886
Grok 4.7 v1.2; xhigh Cursor 31.8% $4.56 100,474
Claude Sonnet 5 v1.2; max 31.3% $4.74 88,420
GPT-5.6 Luna v1.2; xhigh 31.3% $0.57 22,738
Claude Opus 4.8 v1.2; xhigh 30.8% $5.63 52,812
Grok 4.5 v1.2; medium Cursor 30.8% $0.96 20,874
GLM-5.3 v1.2; max 30.8% $2.25 82,968
Claude Opus 5.5 v1.2; low 30.8% $0.24 4,621
Claude Sonnet 5 v1.2; high 30.3% $2.31 36,066
DeepSeek V4.1 Flash 1017 v1.2; minimal 30.3% $0.07 27,858
GPT-5.6 Terra v1.2; medium 29.7% $0.25 6,352
Grok 4.6 v1.2; medium Cursor 29.2% $1.02 22,334
DeepSeek V4.1 Flash 1017 v1.2; max 29.2% $0.09 36,077
Grok 4.6 v1.2; low Cursor 28.7% $0.66 14,500
Claude Opus 4.8 v1.2; medium 28.2% $3.41 24,992
GPT-5.6 Terra v1.2; low 28.2% $0.21 5,059
Claude Opus 4.8 v1.2; high 27.7% $4.08 30,817
Grok 4.7 v1.2; high Cursor 27.7% $3.46 78,044
GLM-5.2 v1.2; max 27.2% $1.23 65,046
Muse Spark 1.1 v1.2; max 27.2% $2.69 153,915
Claude Opus 5 v1.2; low 27.2% $1.19 12,809
Gemini 3.1 Pro v1.2; high 25.6% $0.80 15,429
Muse Spark 1.1 v1.2; high 25.6% $2.21 154,040
Muse Spark 1.1 v1.2; medium 25.1% $1.30 81,463
Muse Spark 1.1 v1.2; xhigh 25.1% $2.57 149,157
Kimi K2.7 Code v1.2; mini-swe-agent-preinstalled mini-swe-agent 23.6% $1.27 45,199
Muse Spark 1.2 v1.2; xhigh 23.6% $4.67 217,019
Gemini 3.1 Pro v1.2; xhigh 23.1% $0.80 16,282
Gemini 3.5 Flash v1.2; medium 23.1% $1.40 39,107
Gemini 3.5 Flash v1.2; xhigh 22.6% $1.44 41,176
Claude Sonnet 5 v1.2; medium 22.6% $1.44 22,228
Gemini 3.1 Pro v1.2; max 22.1% $0.84 17,164
Claude Opus 4.8 v1.2; low 22.1% $1.84 16,746
Gemini 3.1 Pro v1.2; medium 21.5% $0.84 18,624
Gemini 3.5 Flash v1.2; high 21.5% $1.43 39,693
Gemini 3.5 Flash v1.2; low 21.5% $0.92 26,754
DeepSeek V4 Flash 0731 v1.2; max 21.5% $0.26 59,608
GPT-5.6 Luna v1.2; medium 20% $0.15 6,311
Muse Spark 1.1 v1.2; low 17.9% $0.65 37,716
Composer 2.5 v1.2; cursor-cli 17.4% $0.14 6,977
Claude Sonnet 5 v1.2; low 16.9% $0.74 13,080
Gemini 3.1 Pro v1.2; low 12.8% $0.37 8,145
GPT-5.6 Luna v1.2; low 10.8% $0.07 2,865
Inkling-Small v1.2; thinking-0.99 7.69% $0.10 11,504

4 models have held number one on the official leaderboard since its release in July 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic OpenAI
Own leaderboard 4 leaders
Claude Opus 5.5 54.4%

Source: Official ReactBench leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 Claude Fable 5.1 Claude Fable 5 Claude Opus 4.8 Gemini 3.1 Pro 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 GPT-5.6 Sol

Source: Official ReactBench leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Coding

All 282 in Coding
  • SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
  • LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests. DeepSeek-V4-Pro 93.5%
  • SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Fable 5.2 96.6%
  • DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
  • SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases. Seed2.1 Pro 59.8%
  • Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance. DeepSeek V4.1 Flash 3,471