Benchmark Almanac

Vibe Code Bench 1-100

Muse Wonder Full 34% leads since Oct 2026 Not in model cards

Agents start from a working web app built for Vibe Code Bench v1.1. They extend it through up to ten dependent plain-English product requests without breaking it.

Later turns keep the code, data and conversation from earlier work. 26 models on 2026-10-29, led by Wonder Full at 33.98%.

Made by Vals AI Released Sep 2026 Building whole projects Tasks kept private Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
Expert Specialists with years of training, often over hours of work
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
Leaderboard
26
Models scored on its own leaderboard

Muse Wonder Full leads with 34%

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

26 models. Scroll sideways for the rest.

  1. 34.0 Muse Wonder Full
  2. 30.4 Claude Opus 5.5
  3. 28.5 Claude Opus 5
  4. 28.0 Claude Fable 5.1
  5. 27.6 GPT-6 Astra
  6. 26.6 Muse Wonder
  7. 24.8 Qwen3.8 Max (0902)
  8. 22.6 GPT-5.6 Luna
  9. 21.2 Gemini 4 Argon
  10. 20.5 Muse Spark 1.3
  11. 20.4 Claude Haiku 5.5
  12. 20.0 GPT-5.6 Sol
  13. 20.0 GLM-5.3
  14. 18.8 Gemini 3.8 Flash
  15. 18.2 Kimi K3
  16. 17.5 DeepSeek V4 Pro 0813
  17. 16.4 DeepSeek V4.1 Flash
  18. 16.0 GLM-5.3-Flash
  19. 14.8 GPT-5.6 Terra
  20. 14.8 Grok 4.6
  21. 14.3 Mistral Large 4
  22. 13.8 Claude Sonnet 5
  23. 12.8 Qwen 3.8 Max
  24. 9.22 MiniMax-M3
  25. 7.33 Inkling
  26. 6.69 Gemini 3.1 Pro
  1. 1 Muse Wonder Full 34%
  2. 2 Claude Opus 5.5 30.4%
  3. 3 Claude Opus 5 28.5%
  4. 4 Claude Fable 5.1 28%
  5. 5 GPT-6 Astra 27.6%
  6. 6 Muse Wonder 26.6%
  7. 7 Qwen3.8 Max (0902) 24.8%
  8. 8 GPT-5.6 Luna 22.6%
  9. 9 Gemini 4 Argon 21.2%
  10. 10 Muse Spark 1.3 max 20.5%
  11. 11 Claude Haiku 5.5 20.4%
  12. 12 GPT-5.6 Sol 20%
  13. 13 GLM-5.3 20%
  14. 14 Gemini 3.8 Flash 18.8%
  15. 15 Kimi K3 18.2%
  16. 16 DeepSeek V4 Pro 0813 17.5%
  17. 17 DeepSeek V4.1 Flash 16.4%
  18. 18 GLM-5.3-Flash 16%
  19. 19 GPT-5.6 Terra 14.8%
  20. 20 Grok 4.6 14.8%
  21. 21 Mistral Large 4 14.3%
  22. 22 Claude Sonnet 5 13.8%
  23. 23 Qwen 3.8 Max 12.8%
  24. 24 MiniMax-M3 9.22%
  25. 25 Inkling 7.33%
  26. 26 Gemini 3.1 Pro 6.69%
#ModelScoreSettingDateSource
1 Muse Wonder Full 34% Oct 2026 Official leaderboard
2 Claude Opus 5.5 30.4% Sep 2026 Official leaderboard
3 Claude Opus 5 28.5% Jul 2026 Official leaderboard
4 Claude Fable 5.1 28% Sep 2026 Official leaderboard
5 GPT-6 Astra 27.6% Sep 2026 Official leaderboard
6 Muse Wonder 26.6% Oct 2026 Official leaderboard
7 Qwen3.8 Max (0902) 24.8% Sep 2026 Official leaderboard
8 GPT-5.6 Luna 22.6% Jul 2026 Official leaderboard
9 Gemini 4 Argon 21.2% Sep 2026 Official leaderboard
10 Muse Spark 1.3 20.5% max Sep 2026 Official leaderboard
11 Claude Haiku 5.5 20.4% Oct 2026 Official leaderboard
12 GPT-5.6 Sol 20% Jul 2026 Official leaderboard
13 GLM-5.3 20% Aug 2026 Official leaderboard
14 Gemini 3.8 Flash 18.8% Sep 2026 Official leaderboard
15 Kimi K3 18.2% Jul 2026 Official leaderboard
16 DeepSeek V4 Pro 0813 17.5% Aug 2026 Official leaderboard
17 DeepSeek V4.1 Flash 16.4% Sep 2026 Official leaderboard
18 GLM-5.3-Flash 16% Aug 2026 Official leaderboard
19 GPT-5.6 Terra 14.8% Jul 2026 Official leaderboard
20 Grok 4.6 14.8% Aug 2026 Official leaderboard
21 Mistral Large 4 14.3% Oct 2026 Official leaderboard
22 Claude Sonnet 5 13.8% Jun 2026 Official leaderboard
23 Qwen 3.8 Max 12.8% Aug 2026 Official leaderboard
24 MiniMax-M3 9.22% Jun 2026 Official leaderboard
25 Inkling 7.33% Jul 2026 Official leaderboard
26 Gemini 3.1 Pro 6.69% Feb 2026 Official leaderboard

The best score, 34%, costs $20 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Z.ai Qwen DeepSeek Meta Cost frontier One model at each effort level
0% 10% 20% 30% 40% $0.10 $1.00 $10 $100 $1,000 Cost per task, log scale Muse Wonder Full Muse Wonder Qwen3.8 Max (0902) GPT-5.6 Luna DeepSeek V4.1 Flash 0% 10% 20% 30% 40% $0.10 $1.00 $10 $100 $1,000 Cost per task, log scale Muse Wonder Full Muse Wonder
ModelEffortHarnessScoreCost per taskOutput tokens
Muse Wonder Full 34% $20
Claude Opus 5.5 30.4% $44
Claude Opus 5 28.5% $41
Claude Fable 5.1 28% $150
GPT-6 Astra 27.6% $85
Muse Wonder 26.6% $14
Qwen3.8 Max (0902) 24.8% $12
GPT-5.6 Luna 22.6% $4.91
Gemini 4 Argon 21.2% $19
Muse Spark 1.3 max 20.5% $5.60
Claude Haiku 5.5 20.4% $8.30
GPT-5.6 Sol 20% $47
GLM-5.3 20% $8.24
Gemini 3.8 Flash 18.8% $19
Kimi K3 18.2% $9.40
DeepSeek V4 Pro 0813 17.5% $5.89
DeepSeek V4.1 Flash 16.4% $0.69
GLM-5.3-Flash 16% $0.46
GPT-5.6 Terra 14.8% $33
Grok 4.6 14.8% $26
Mistral Large 4 14.3% $3.44
Claude Sonnet 5 13.8% $71
Qwen 3.8 Max 12.8% $9.68
MiniMax-M3 9.22% $3.03
Inkling 7.33% $5.00
Gemini 3.1 Pro 6.69% $3.36

3 models have held number one since its release in Sep 2026

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

Anthropic Meta
Own leaderboard 3 leaders
Muse Wonder Full 34%

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%) Muse Wonder Full Claude Opus 5 GPT-5.6 Luna Claude Sonnet 5 MiniMax-M3 Gemini 3.1 Pro 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Muse Wonder Full GPT-5.6 Luna

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

More in Coding

All 243 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%
  • SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Opus 5.5 93.9%
  • 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%
  • 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