CyberBench (Vals AI) leaderboard
- Muse Wonder leads
- 80.4%
- models measured
- 41
- lead over #2
- +2.4 pts
- median
- 69.5%
Leader on CyberBench (Vals AI) · read October 21
# CyberBench (Vals AI), read October 21
leaders = {
"Muse Wonder": 80.4%, "GPT-6 Sol": 78.0%, "Gemini 4 Argon": 77.9%,
}
On October 21, 2026, 41 of our catalog's models had a CyberBench (Vals AI) result, and Meta Muse Wonder led with 80.4%.
Agents find and patch security vulnerabilities. The accuracy Vals AI publishes for it.
Role in the index
A coding board we show beside the Vibe Coding Index, not inside it: the index keeps the four boards it was defined on, so a score does not move every time a board is added. Read the methodology
| # | Model | Result | Cost per test | Duration, as printed | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|---|
| 1 | Muse WonderMeta | 80.4% | $3.66 | 13m 45s | 64.5 | $20.00 | |
| 2 | GPT-6 SolOpenAI | 78.0% | $1.34 | 12m 14s | 82.8 | $10.00 | |
| 3 | Gemini 4 ArgonGoogle | 77.9% | $5.11 | 19m 10s | 80.5 | $10.00 | |
| 4 | GPT-5.6 SolOpenAI | 76.3% | $3.05 | 10m 47s | 76.7 | $20.00 | |
| 5 | GPT-6 LunaOpenAI | 76.3% | $0.13 | 15m 50s | 71.4 | $0.50 | |
| 6 | Claude Haiku 5.5Anthropic | 75.5% | $1.25 | 32m 39s | 75.9 | $0.50 | |
| 7 | MiMo-V2.6-FlashXiaomi | 75.4% | $0.05 | 28m 22s | 61.9 | $0.28 | |
| 8 | Kimi K3Moonshot AI | 75.2% | $2.13 | 30m 44s | 61.8 | $11.936 | |
| 9 | Claude Opus 5.5Anthropic | 74.6% | $4.17 | 23m 02s | 92.2 | $20.00 | |
| 10 | DeepSeek V4.1 FlashDeepSeek | 73.7% | $0.07 | 13m 50s | 77.2 | $0.525 | |
| 11 | GPT-5.6 LunaOpenAI | 73.6% | $0.40 | 16m 15s | 61.5 | $1.20 | |
| 12 | DeepSeek V4 Pro 0813DeepSeek | 72.9% | $0.59 | 22m 33s | 54.6 | $1.98 | |
| 13 | MiMo-V2.6-ProXiaomi | 72.9% | $0.09 | 27m 41s | 72.6 | $0.87 | |
| 14 | DeepSeek V4 Flash 0731DeepSeek | 72.1% | $0.19 | 15m 28s | Not benchmarked | $1.32 | |
| 15 | GPT-5.6 TerraOpenAI | 72.1% | $3.31 | 17m 04s | 69.0 | $12.00 | |
| 16 | GLM 5.3Z.ai | 72.1% | $2.69 | 30m 18s | 73.1 | $1.40 | |
| 17 | Grok 4.5xAI | 71.0% | $2.95 | 14m 08s | 75.5 | $6.00 | |
| 18 | Claude Fable 5.1Anthropic | 70.4% | $4.01 | 18m 32s | 89.1 | $50.00 | |
| 19 | Claude Sonnet 5Anthropic | 70.3% | $4.84 | 22m 57s | 63.5 | $10.00 | |
| 20 | Ember-1Fireworks | 69.5% | $2.43 | 22m 02s | Not benchmarked | $15.00 | |
| 21 | Muse Spark 1.2Meta | 69.5% | $2.15 | 20m 21s | 65.1 | $4.25 | |
| 22 | Grok 4.7xAI | 69.5% | $8.54 | 15m 30s | 75.0 | $6.00 | |
| 23 | Muse Spark 1.3Meta | 69.4% | $4.26 | 24m 53s | 79.7 | $4.25 | |
| 24 | GLM 5.3 FlashZ.ai | 67.9% | $0.14 | 35m 48s | 71.0 | $0.50 | |
| 25 | Qwen3.8 27BQwen | 67.8% | $0.99 | 41m 36s | 73.6 | $1.35 | |
| 26 | Grok 4.6xAI | 67.7% | $4.07 | 36m 52s | 76.4 | $6.00 | |
| 27 | T InklingThinking Machines | 66.8% | $1.46 | 13m 31s | 49.7 | $4.05 | |
| 28 | Claude Opus 5Anthropic | 65.4% | $2.87 | 19m 25s | 81.8 | $25.00 | |
| 29 | Hy4 previewTencent | 65.4% | $0.49 | 36m 43s | 61.6 | $2.501 | |
| 30 | MiniMax M3MiniMax | 65.0% | $3.41 | 40m 56s | 57.8 | $1.20 | |
| 31 | T Inkling SmallThinking Machines | 64.1% | $0.25 | 19m 35s | 29.9 | $1.20 | |
| 32 | Claude Sonnet 5.5Anthropic | 59.6% | $4.42 | 33m 53s | 88.9 | $10.00 | |
| 33 | Gemini 3.5 Flash LiteGoogle | 59.1% | $0.43 | 17m 56s | 33.6 | $2.50 | |
| 34 | Mercury 2.5Inception | 45.5% | $0.12 | 18m 59s | Not benchmarked | $0.15 | |
| 35 | Gemini 3.6 FlashGoogle | 45.4% | $1.31 | 11m 00s | 48.3 | $3.75 | |
| 36 | Gemini 3.7 FlashGoogle | 43.8% | $2.15 | 7m 34s | 69.9 | $3.75 | |
| 37 | Gemini 3.8 FlashGoogle | 43.8% | $1.31 | 9m 49s | 70.5 | $4.125 | |
| 38 | GPT-6 AstraOpenAI | 41.1% | $1.84 | 4m 10s | 88.7 | $50.00 | |
| 39 | GPT-6.1 SolOpenAI | 39.3% | $0.29 | 3m 42s | 88.7 | $10.00 | |
| 40 | Nemotron 3.5 LightningNVIDIA | 29.5% | $0.09 | 50m 32s | Not benchmarked | $0.20 | |
| 41 | Qwen3.8 Max (0902)Qwen | 28.6% | $0.78 | 9m 40s | 63.5 | $6.00 |
Only models in our catalog are listed, and each appears once with its best published run. 132 rows were read from the source on October 21, 2026, and rows that match no catalog model are left out. Cost per test is what Vals AI reports one test cost on the run behind the result, in US dollars. Every other column is a figure the board publishes for the same run.