Terminal-Bench 2.1 (Vals AI) leaderboard
- Claude Opus 5.5 leads
- 87.6%
- models measured
- 68
- lead over #2
- +0.4 pts
- median
- 67.0%
Leader on Terminal-Bench 2.1 (Vals AI) · read October 22
# Terminal-Bench 2.1 (Vals AI), read October 22
leaders = {
"Claude Opus 5.5": 87.6%, "GPT-6 Astra": 87.3%, "GPT-5.6 Sol": 85.8%,
}
On October 22, 2026, 68 of our catalog's models had a Terminal-Bench 2.1 (Vals AI) result, and Anthropic Claude Opus 5.5 led with 87.6%, ahead of OpenAI GPT-6 Astra at 87.3%.
Difficult tasks finished in a real terminal, run by Vals AI with one shared harness. Share completed.
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 | Claude Opus 5.5Anthropic | 87.6% | $0.50 | 6m 23s | 88.8 | $20.00 | |
| 2 | GPT-6 AstraOpenAI | 87.3% | $1.34 | 4m 34s | 85.1 | $50.00 | |
| 3 | GPT-5.6 SolOpenAI | 85.8% | $1.02 | 6m 13s | 73.5 | $20.00 | |
| 4 | Claude Fable 5.1Anthropic | 85.0% | $3.07 | 8m 20s | 84.2 | $50.00 | |
| 5 | Claude Opus 5Anthropic | 84.6% | $0.89 | 12m 33s | 83.0 | $25.00 | |
| 6 | Claude Sonnet 5.5Anthropic | 83.2% | $0.63 | 7m 46s | 87.3 | $10.00 | |
| 7 | GPT-6 SolOpenAI | 83.2% | $0.37 | 4m 59s | 76.0 | $10.00 | |
| 8 | Gemini 3.8 FlashGoogle | 81.3% | $1.54 | 9m 15s | 61.6 | $3.75 | |
| 9 | Kimi K3Moonshot AI | 80.9% | $0.34 | 10m 09s | 62.3 | $15.00 | |
| 10 | Claude Fable 5Anthropic | 80.5% | $1.43 | 8m 25s | 79.3 | $50.00 | |
| 11 | GPT-5.6 LunaOpenAI | 79.0% | $0.05 | 5m 56s | 58.0 | $1.20 | |
| 12 | Grok 4.6xAI | 78.3% | $0.45 | 10m 43s | 65.7 | $6.00 | |
| 13 | Gemini 3.7 FlashGoogle | 77.5% | $1.02 | 8m 12s | 58.0 | $3.75 | |
| 14 | GPT-5.6 TerraOpenAI | 77.5% | $0.47 | 8m 40s | 63.9 | $12.00 | |
| 15 | GPT-5.5OpenAI | 76.4% | $0.74 | 7m 07s | 71.3 | $30.00 | |
| 16 | MiMo-V2.6-FlashXiaomi | 76.4% | $0.02 | 13m 30s | 54.7 | $0.28 | |
| 17 | Claude Sonnet 5Anthropic | 74.5% | $0.53 | 10m 35s | 57.7 | $10.00 | |
| 18 | DeepSeek V4.1 FlashDeepSeek | 74.5% | $0.10 | 10m 05s | 62.5 | $0.84 | |
| 19 | Gemini 3.5 FlashGoogle | 74.2% | $0.87 | 6m 13s | 39.8 | $9.00 | |
| 20 | Gemini 3.6 FlashGoogle | 73.8% | $1.17 | 9m 12s | 62.5 | $3.75 | |
| 21 | Grok 4.7xAI | 73.4% | $1.08 | 13m 18s | 66.7 | $6.00 | |
| 22 | GPT-6 LunaOpenAI | 73.0% | $0.03 | 7m 05s | 57.3 | $0.50 | |
| 23 | Muse Spark 1.3Meta | 72.3% | $0.66 | 14m 18s | 57.4 | $4.25 | |
| 24 | Claude Opus 4.8Anthropic | 71.9% | $2.41 | 15m 30s | 65.0 | $25.00 | |
| 25 | GLM 5.3Z.ai | 71.5% | $0.31 | 13m 07s | 70.6 | $4.40 | |
| 26 | Gemini 3.1 Pro PreviewGoogle | 70.8% | $0.58 | 6m 50s | 32.3 | $12.00 | |
| 27 | Muse Spark 1.2Meta | 69.7% | $0.50 | 14m 39s | 55.7 | $4.25 | |
| 28 | Muse Spark 1.1Meta | 69.3% | $0.42 | 10m 25s | 70.3 | $4.25 | |
| 29 | Claude Opus 4.7Anthropic | 68.5% | $1.98 | 13m 37s | 67.5 | $25.00 | |
| 30 | Grok 4.5xAI | 67.8% | $0.28 | 10m 24s | 58.6 | $6.00 | |
| 31 | MiMo-V2.6-ProXiaomi | 67.8% | $0.07 | 15m 39s | 65.4 | $0.87 | |
| 32 | GLM 5.2Z.ai | 67.8% | $0.43 | 13m 41s | 61.1 | $4.40 | |
| 33 | Qwen3.8 Max (0902)Qwen | 67.4% | $0.45 | 14m 51s | 62.4 | $6.00 | |
| 34 | DeepSeek V4 Flash 0731DeepSeek | 67.0% | $0.02 | 11m 23s | 51.8 | $0.396 | |
| 35 | Kimi K2.7 CodeMoonshot AI | 67.0% | $0.26 | 13m 04s | 53.4 | $4.00 | |
| 36 | GLM 5.3 FlashZ.ai | 62.9% | $0.03 | 16m 06s | 57.7 | $0.50 | |
| 37 | Qwen3.7 MaxQwen | 61.0% | $0.33 | 15m 35s | 54.1 | $4.425 | |
| 38 | MiMo-V2.5Xiaomi | 60.7% | $0.02 | 12m 55s | 50.0 | $0.336 | |
| 39 | Qwen3.8 27BQwen | 58.4% | $0.68 | 14m 53s | 74.2 | $2.55 | |
| 40 | Gemini 4 ArgonGoogle | 57.7% | Not published | 14m 02s | 81.0 | $10.00 | |
| 41 | Claude Sonnet 4.6Anthropic | 57.3% | $0.57 | 11m 44s | 57.3 | $15.00 | |
| 42 | MiMo-V2.5-ProXiaomi | 57.3% | $0.04 | 14m 26s | 31.6 | $0.9152 | |
| 43 | GLM 5.1Z.ai | 56.9% | $0.23 | 15m 41s | 49.2 | $4.40 | |
| 44 | Hy4 previewTencent | 55.1% | $0.14 | 18m 29s | 44.3 | $2.501 | |
| 45 | T Inkling SmallThinking Machines | 55.1% | $0.13 | 10m 04s | 27.3 | $1.20 | |
| 46 | DeepSeek V4 Pro 0813DeepSeek | 54.7% | $0.20 | 14m 22s | 56.1 | $1.98 | |
| 47 | GPT-5.4 MiniOpenAI | 54.7% | $0.29 | 11m 13s | 35.8 | $4.50 | |
| 48 | Gemini 3 Flash PreviewGoogle | 53.9% | $0.15 | 6m 16s | 43.7 | $3.00 | |
| 49 | MiniMax M3MiniMax | 53.6% | $0.21 | 17m 03s | 39.2 | $1.20 | |
| 50 | Kimi K2.6Moonshot AI | 53.6% | $0.14 | 15m 32s | 52.1 | $4.00 | |
| 51 | Qwen3.6 PlusQwen | 53.2% | $0.20 | 12m 58s | 44.1 | $1.95 | |
| 52 | Qwen3.7 PlusQwen | 52.8% | $0.10 | 10m 55s | 33.2 | $1.28 | |
| 53 | Nemotron 3 UltraNVIDIA | 50.9% | Not published | 9m 44s | 30.0 | $2.40 | |
| 54 | DeepSeek V4 Pro 0423DeepSeek | 50.2% | $0.24 | 17m 58s | 38.7 | $2.976 | |
| 55 | Gemini 3.5 Flash LiteGoogle | 50.2% | $0.34 | 6m 56s | 49.0 | $2.50 | |
| 56 | Ling 3.0 FlashinclusionAI | 50.2% | $0.07 | 9m 54s | 29.6 | $0.1215 | |
| 57 | MiniMax M2.7MiniMax | 48.7% | $0.07 | 13m 33s | 36.2 | $1.20 | |
| 58 | T InklingThinking Machines | 47.6% | $0.63 | 12m 52s | 31.5 | $4.05 | |
| 59 | Grok 4.20xAI | 44.2% | $0.41 | 5m 06s | 33.3 | $2.50 | |
| 60 | Claude Haiku 4.5Anthropic | 43.8% | $0.36 | 10m 49s | 35.6 | $5.00 | |
| 61 | Kimi K2.5Moonshot AI | 42.0% | $0.09 | 17m 31s | 38.6 | $2.85 | |
| 62 | Grok 4.3xAI | 42.0% | $0.98 | 7m 51s | 37.4 | $2.50 | |
| 63 | GPT-5.4 NanoOpenAI | 41.6% | $0.07 | 12m 37s | 44.8 | $1.25 | |
| 64 | Mistral Medium 3.5Mistral AI | 39.0% | $1.19 | 16m 29s | 25.5 | $7.50 | |
| 65 | Mercury 2.5Inception | 34.5% | $0.21 | 11m 29s | 1.6 | $0.15 | |
| 66 | Gemini 3.1 Flash Lite PreviewGoogle | 34.1% | $0.08 | 4m 38s | 22.7 | $1.50 | |
| 67 | Command ACohere | 17.6% | Not published | Not published | Not benchmarked | $10.00 | |
| 68 | Nemotron 3.5 LightningNVIDIA | 10.9% | $0.01 | 17m 08s | 22.6 | $0.147 |
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.