Benchmark Almanac

CyberBench v1.1 (Vals)

Muse Wonder 80.4% leads since Oct 2026 Not in model cards

Agents work on OSS-Fuzz memory-safety bugs from ARVO images in two tracks. They write a proof-of-concept input that crashes the target, or patch the source so the crash stops and tests still pass.

Vals reran the benchmark with offline task sandboxes, a hardened clean-room grader and a new Patch roster, and archived v1. Release month is inferred: v1 last updated 2026-09-16, v1.1 shows 2026-10-20.

Made by Vals AI Released Oct 2026 Finding and exploiting real bugs 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 vals.ai
46
Entries on its own leaderboard, 43 models we track

Muse Wonder leads with 80.4%

Best score per entry 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. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

46 entries. Scroll sideways for the rest.

  1. 80.4 Muse Wonder
  2. 78.0 GPT-6 Sol
  3. 77.9 Gemini 4 Argon
  4. 76.3 GPT-5.6 Sol
  5. 76.3 GPT-6 Luna
  6. 75.5 Claude Haiku 5.5
  7. 75.4 MiMo-V2.6 Flash
  8. 75.2 Kimi K3
  9. 74.6 Claude Opus 5.5 (CVP)
  10. 73.7 DeepSeek V4.1 Flash
  11. 73.6 GPT-5.6 Luna
  12. 73.6 Muse Wonder Full
  13. 73.0 Daybreak Blue
  14. 72.9 DeepSeek V4 Pro 0813
  15. 72.9 MiMo-V2.6 Pro
  16. 72.7 Muse Spark 1.3
  17. 72.1 GPT-5.6 Terra
  18. 72.1 DeepSeek V4 Flash 0731
  19. 72.1 GLM-5.3
  20. 71.0 Grok 4.5
  21. 70.4 Claude Fable 5.1
  22. 70.3 Claude Sonnet 5 (Cyber Range)
  23. 69.5 Muse Spark 1.2
  24. 69.5 Ember-1
  25. 69.5 Grok 4.7
  26. 67.9 GLM-5.3-Flash
  27. 67.8 Qwen 3.8 27B
  28. 67.7 Grok 4.6
  29. 66.8 Inkling
  30. 65.4 Claude Opus 5
  31. 65.4 Tencent Hy4 preview
  32. 65.0 MiniMax-M3
  33. 64.1 Inkling-Small
  34. 61.9 Claude Sonnet 5
  35. 59.6 Claude Sonnet 5.5
  36. 59.1 Gemini 3.5 Flash-Lite
  37. 58.3 Ling-3.0-flash-Fin
  38. 55.4 Claude Opus 5.5
  39. 45.5 Mercury 2.5
  40. 45.4 Gemini 3.6 Flash
  41. 43.8 Gemini 3.7 Flash
  42. 43.8 Gemini 3.8 Flash
  43. 41.1 GPT-6 Astra
  44. 39.3 GPT-6.1 Sol
  45. 29.5 Nemotron 3.5 Lightning 30B-A3B
  46. 28.6 Qwen 3.8 Max
  1. 1 Muse Wonder 80.4%
  2. 2 GPT-6 Sol 78%
  3. 3 Gemini 4 Argon 77.9%
  4. 4 GPT-5.6 Sol 76.3%
  5. 5 GPT-6 Luna 76.3%
  6. 6 Claude Haiku 5.5 75.5%
  7. 7 MiMo-V2.6 Flash 75.4%
  8. 8 Kimi K3 75.2%
  9. 9 Claude Opus 5.5 (CVP) 74.6%
  10. 10 DeepSeek V4.1 Flash 73.7%
  11. 11 GPT-5.6 Luna 73.6%
  12. 12 Muse Wonder Full 73.6%
  13. 13 Daybreak Blue 73%
  14. 14 DeepSeek V4 Pro 0813 72.9%
  15. 15 MiMo-V2.6 Pro 72.9%
  16. 16 Muse Spark 1.3 max 72.7%
  17. 17 GPT-5.6 Terra 72.1%
  18. 18 DeepSeek V4 Flash 0731 72.1%
  19. 19 GLM-5.3 72.1%
  20. 20 Grok 4.5 71%
  21. 21 Claude Fable 5.1 70.4%
  22. 22 Claude Sonnet 5 (Cyber Range) 70.3%
  23. 23 Muse Spark 1.2 69.5%
  24. 24 Ember-1 69.5%
  25. 25 Grok 4.7 69.5%
  26. 26 GLM-5.3-Flash 67.9%
  27. 27 Qwen 3.8 27B 67.8%
  28. 28 Grok 4.6 67.7%
  29. 29 Inkling 66.8%
  30. 30 Claude Opus 5 65.4%
  31. 31 Tencent Hy4 preview 65.4%
  32. 32 MiniMax-M3 65%
  33. 33 Inkling-Small 64.1%
  34. 34 Claude Sonnet 5 61.9%
  35. 35 Claude Sonnet 5.5 59.6%
  36. 36 Gemini 3.5 Flash-Lite 59.1%
  37. 37 Ling-3.0-flash-Fin 58.3%
  38. 38 Claude Opus 5.5 55.4%
  39. 39 Mercury 2.5 45.5%
  40. 40 Gemini 3.6 Flash 45.4%
  41. 41 Gemini 3.7 Flash 43.8%
  42. 42 Gemini 3.8 Flash 43.8%
  43. 43 GPT-6 Astra 41.1%
  44. 44 GPT-6.1 Sol 39.3%
  45. 45 Nemotron 3.5 Lightning 30B-A3B 29.5%
  46. 46 Qwen 3.8 Max 28.6%
#EntryScoreSettingDateSource
1 Muse Wonder 80.4% Oct 2026 Official leaderboard
2 GPT-6 Sol 78% Sep 2026 Official leaderboard
3 Gemini 4 Argon 77.9% Sep 2026 Official leaderboard
4 GPT-5.6 Sol 76.3% Jul 2026 Official leaderboard
5 GPT-6 Luna 76.3% Sep 2026 Official leaderboard
6 Claude Haiku 5.5 75.5% Oct 2026 Official leaderboard
7 MiMo-V2.6 Flash 75.4% Sep 2026 Official leaderboard
8 Kimi K3 75.2% Jul 2026 Official leaderboard
9 Claude Opus 5.5 (CVP) 74.6% Official leaderboard
10 DeepSeek V4.1 Flash 73.7% Sep 2026 Official leaderboard
11 GPT-5.6 Luna 73.6% Jul 2026 Official leaderboard
12 Muse Wonder Full 73.6% Oct 2026 Official leaderboard
13 Daybreak Blue 73% Official leaderboard
14 DeepSeek V4 Pro 0813 72.9% Aug 2026 Official leaderboard
15 MiMo-V2.6 Pro 72.9% Sep 2026 Official leaderboard
16 Muse Spark 1.3 72.7% max Sep 2026 Official leaderboard
17 GPT-5.6 Terra 72.1% Jul 2026 Official leaderboard
18 DeepSeek V4 Flash 0731 72.1% Jul 2026 Official leaderboard
19 GLM-5.3 72.1% Aug 2026 Official leaderboard
20 Grok 4.5 71% Jul 2026 Official leaderboard
21 Claude Fable 5.1 70.4% Sep 2026 Official leaderboard
22 Claude Sonnet 5 (Cyber Range) 70.3% Official leaderboard
23 Muse Spark 1.2 69.5% Aug 2026 Official leaderboard
24 Ember-1 69.5% Sep 2026 Official leaderboard
25 Grok 4.7 69.5% Sep 2026 Official leaderboard
26 GLM-5.3-Flash 67.9% Aug 2026 Official leaderboard
27 Qwen 3.8 27B 67.8% Aug 2026 Official leaderboard
28 Grok 4.6 67.7% Aug 2026 Official leaderboard
29 Inkling 66.8% Jul 2026 Official leaderboard
30 Claude Opus 5 65.4% Jul 2026 Official leaderboard
31 Tencent Hy4 preview 65.4% Aug 2026 Official leaderboard
32 MiniMax-M3 65% Jun 2026 Official leaderboard
33 Inkling-Small 64.1% Jul 2026 Official leaderboard
34 Claude Sonnet 5 61.9% Jun 2026 Official leaderboard
35 Claude Sonnet 5.5 59.6% Sep 2026 Official leaderboard
36 Gemini 3.5 Flash-Lite 59.1% Jul 2026 Official leaderboard
37 Ling-3.0-flash-Fin 58.3% Sep 2026 Official leaderboard
38 Claude Opus 5.5 55.4% Sep 2026 Official leaderboard
39 Mercury 2.5 45.5% Sep 2026 Official leaderboard
40 Gemini 3.6 Flash 45.4% Jul 2026 Official leaderboard
41 Gemini 3.7 Flash 43.8% Aug 2026 Official leaderboard
42 Gemini 3.8 Flash 43.8% Sep 2026 Official leaderboard
43 GPT-6 Astra 41.1% Sep 2026 Official leaderboard
44 GPT-6.1 Sol 39.3% Sep 2026 Official leaderboard
45 Nemotron 3.5 Lightning 30B-A3B 29.5% Aug 2026 Official leaderboard
46 Qwen 3.8 Max 28.6% Aug 2026 Official leaderboard

The best score, 80.4%, costs $3.66 a task

Runs on the cost frontier score higher than any run that's cheaper. MiMo-V2.6 Flash comes within 5 points of the best for $0.05 a task, 73 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Xiaomi Corp OpenAI Meta Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale Muse Wonder GPT-6 Sol GPT-6 Luna MiMo-V2.6 Flash 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale Muse Wonder GPT-6 Luna
ModelEffortHarnessScoreCost per taskOutput tokens
Muse Wonder 80.4% $3.66
GPT-6 Sol 78% $1.34
Gemini 4 Argon 77.9% $5.11
GPT-5.6 Sol 76.3% $3.05
GPT-6 Luna 76.3% $0.13
Claude Haiku 5.5 75.5% $1.25
MiMo-V2.6 Flash 75.4% $0.05
Kimi K3 75.2% $2.13
DeepSeek V4.1 Flash 73.7% $0.11
GPT-5.6 Luna 73.6% $0.40
Muse Wonder Full 73.6% $3.63
DeepSeek V4 Pro 0813 72.9% $0.59
MiMo-V2.6 Pro 72.9% $0.09
Muse Spark 1.3 max 72.7% $3.34
GPT-5.6 Terra 72.1% $3.31
DeepSeek V4 Flash 0731 72.1% $0.19
GLM-5.3 72.1% $2.69
Grok 4.5 71% $2.95
Claude Fable 5.1 70.4% $4.01
Muse Spark 1.2 69.5% $2.15
Ember-1 69.5% $2.43
Grok 4.7 69.5% $8.54
Muse Spark 1.3 69.4% $4.26
GLM-5.3-Flash 67.9% $0.14
Qwen 3.8 27B 67.8% $0.99
Grok 4.6 67.7% $4.07
Inkling 66.8% $1.46
Claude Opus 5 65.4% $3.53
Tencent Hy4 preview 65.4% $0.46
MiniMax-M3 65% $3.41
Inkling-Small 64.1% $0.25
Claude Sonnet 5 61.9% $1.89
Claude Sonnet 5.5 59.6% $4.42
Gemini 3.5 Flash-Lite 59.1% $0.43
Ling-3.0-flash-Fin 58.3% $0.10
Claude Opus 5.5 55.4% $3.48
Mercury 2.5 45.5% $0.12
Gemini 3.6 Flash 45.4% $1.31
Gemini 3.7 Flash 43.8% $2.15
Gemini 3.8 Flash 43.8% $1.96
GPT-6 Astra 41.1% $1.84
GPT-6.1 Sol 39.3% $0.29
Nemotron 3.5 Lightning 30B-A3B 29.5% $0.09
Qwen 3.8 Max 28.6% $0.78

2 models have held number one since its release in Oct 2026

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

OpenAI Meta
Own leaderboard 2 leaders
Muse Wonder 80.4%

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 GPT-5.6 Sol Grok 4.5 MiniMax-M3 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Muse Wonder Grok 4.5

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.

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

OpenAI Meta
CyberBench v1.1 (Vals) 4 leaders
Muse Wonder 80.4%
CyberBench (Vals) Passed 85% in Jul 2026, by GPT-5.6 Sol
GPT-5.6 Sol 88.1%
CyberBench v1.1 (Vals) 2 leaders
Muse Wonder 80.4%
Replaces CyberBench (Vals)

More in Cybersecurity

All 111 in Cybersecurity
  • ExploitGym In use 869 real-vulnerability exploit-development tasks (userspace, V8, Linux kernel). Success means capturing a flag through the intended vulnerability. GPT-6 Astra 42.4%
  • ExploitBench In use 41 recent V8 JavaScript-engine N-day vulnerabilities. Progress measured by 16 grader-verified capability flags across five tiers. GPT-6 Astra 100%
  • CWE-bench In use Agents audit a repository and write a patch for a security flaw, without being told where the bug is. Covers six programming languages. Gemini 4 Carbon 71%
  • Real-world vulnerability discovery In use Agents search popular open-source codebases to rediscover about 1,200 recent confirmed vulnerabilities across many languages. Reports share discovered. Gemini 4 Argon 86%
  • Artificial Analysis Cyber Index In use Agents do enterprise cyber defense tasks from three benchmarks, CWE-Bench-AA, DeepsecBench-AA and CyberGym-E2E-AA, combined into one success rate. Mistral Large 4 (Le Chonk, open weights) 51%
  • Big Sleep Evaluation (Google) In use Agents search large real codebases such as Chrome and Safari for critical, hard to find vulnerabilities. Scored by pass@1 success rate, run without safety guardrails. Gemini 3.5 Flash Cyber 72%