SkillsBench (Vals AI) leaderboard
- models measured
- 34
- lead over #2
- +3.8 pts
- median
- 52.2%
Leader on SkillsBench (Vals AI) · read October 22
# SkillsBench (Vals AI), read October 22
leaders = {
"DeepSeek V4.1 Flash": 69.8%, "Grok 4.5": 66.0%, "Gemini 3.7 Flash": 65.9%,
}
On October 22, 2026, 34 of our catalog's models had a SkillsBench (Vals AI) result, and DeepSeek V4.1 Flash led with 69.8%, ahead of SpaceXAI Grok 4.5 at 66.0%.
Coding agents given task-specific skills, all run in one agent harness. The score as Vals AI publishes it.
Role in the index
A coding board we show beside the Vibe Coding Index, not inside it: the index keeps the six 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 | DeepSeek V4.1 FlashDeepSeek | 69.8% | $0.09 | 3m 43s | 62.5 | $0.84 | |
| 2 | Grok 4.5xAI | 66.0% | $0.58 | 2m 16s | 58.6 | $6.00 | |
| 3 | Gemini 3.7 FlashGoogle | 65.9% | $1.80 | 3m 06s | 58.0 | $3.75 | |
| 4 | GPT-5.5OpenAI | 62.5% | $1.57 | 4m 40s | 71.3 | $30.00 | |
| 5 | Claude Fable 5.1Anthropic | 61.5% | $4.52 | 11m 02s | 84.2 | $50.00 | |
| 6 | GPT-5.6 LunaOpenAI | 60.5% | $0.21 | 5m 20s | 58.0 | $1.20 | |
| 7 | Claude Opus 5Anthropic | 60.4% | $2.51 | 8m 25s | 83.0 | $25.00 | |
| 8 | Claude Opus 4.8Anthropic | 59.2% | $4.77 | 15m 30s | 65.0 | $25.00 | |
| 9 | Muse Spark 1.1Meta | 59.2% | $0.45 | 6m 16s | 70.3 | $4.25 | |
| 10 | GPT-5.6 TerraOpenAI | 58.9% | $1.75 | 6m 55s | 63.9 | $12.00 | |
| 11 | Gemini 3.8 FlashGoogle | 58.0% | $2.34 | 3m 55s | 61.6 | $3.75 | |
| 12 | Grok 4.6xAI | 55.8% | $0.92 | 13m 28s | 65.7 | $6.00 | |
| 13 | Qwen3.7 PlusQwen | 54.3% | $0.14 | 12m 04s | 33.2 | $1.28 | |
| 14 | GPT-5.6 SolOpenAI | 54.1% | $4.14 | 13m 07s | 73.5 | $20.00 | |
| 15 | DeepSeek V4 Pro 0813DeepSeek | 53.8% | $0.37 | 11m 15s | 56.1 | $1.98 | |
| 16 | Muse Spark 1.2Meta | 53.0% | $0.65 | 13m 00s | 55.7 | $4.25 | |
| 17 | Gemini 3.5 FlashGoogle | 52.7% | $1.25 | 7m 15s | 39.8 | $9.00 | |
| 18 | GPT-5.4OpenAI | 51.7% | $2.04 | 9m 02s | 64.4 | $15.00 | |
| 19 | MiniMax M3MiniMax | 51.5% | $0.50 | 15m 28s | 39.2 | $1.20 | |
| 20 | DeepSeek V4 Pro 0423DeepSeek | 51.3% | $0.40 | 8m 16s | 38.7 | $2.976 | |
| 21 | DeepSeek V4 Flash 0731DeepSeek | 50.7% | $0.13 | 6m 52s | 51.8 | $0.396 | |
| 22 | Kimi K2.7 CodeMoonshot AI | 50.0% | $0.44 | 13m 24s | 53.4 | $4.00 | |
| 23 | Claude Sonnet 4.6Anthropic | 49.0% | $1.38 | 14m 20s | 57.3 | $15.00 | |
| 24 | GLM 5.3Z.ai | 47.5% | $0.71 | 13m 07s | 70.6 | $4.40 | |
| 25 | Claude Sonnet 5Anthropic | 46.5% | $2.90 | 15m 12s | 57.7 | $10.00 | |
| 26 | GLM 5.2Z.ai | 45.1% | $0.81 | 19m 52s | 61.1 | $4.40 | |
| 27 | Qwen3.8 Max (0902)Qwen | 42.0% | $0.52 | 15m 48s | 62.4 | $6.00 | |
| 28 | Grok 4.3xAI | 40.6% | $0.42 | 4m 37s | 37.4 | $2.50 | |
| 29 | GLM 5.3 FlashZ.ai | 40.2% | $0.04 | 12m 09s | 57.7 | $0.50 | |
| 30 | Qwen3.8 27BQwen | 38.1% | $0.82 | 6m 41s | 74.2 | $2.55 | |
| 31 | T Inkling SmallThinking Machines | 33.6% | $0.15 | 11m 41s | 27.3 | $1.20 | |
| 32 | Ling 3.0 FlashinclusionAI | 27.9% | $0.06 | 4m 52s | 29.6 | $0.1215 | |
| 33 | T InklingThinking Machines | 26.1% | $0.83 | 15m 39s | 31.5 | $4.05 | |
| 34 | Mercury 2.5Inception | 18.1% | $0.10 | 49 s | 1.6 | $0.15 |
Only models in our catalog are listed, and each appears once with its best published run. 137 rows were read from the source on October 22, 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.