SREBench (Vals AI) leaderboard
- GPT-6 Astra leads
- 56.9%
- models measured
- 31
- lead over #2
- +6.1 pts
- median
- 3.8%
Leader on SREBench (Vals AI) · read October 22
On October 22, 2026, 31 of our catalog's models had a SREBench (Vals AI) result, and OpenAI GPT-6 Astra led with 56.9%, ahead of OpenAI GPT-6.1 Sol at 50.8%.
Vals AI's site reliability board, run with one shared harness; its page describes the tasks. The score as Vals AI publishes it.
Role in the index
SREBench (Vals AI) is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Cost per test | Duration, as printed | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 56.9% | $10.24 | 17m 51s | 85.1 | $50.00 | |
| 2 | GPT-6.1 SolOpenAI | 50.8% | $2.69 | 32m 32s | 82.4 | $10.00 | |
| 3 | Gemini 4 ArgonGoogle | 44.3% | $30.45 | 1h 35m | 81.0 | $10.00 | |
| 4 | Claude Opus 5.5Anthropic | 33.6% | $34.65 | 1h 37m | 88.8 | $20.00 | |
| 5 | GPT-5.6 SolOpenAI | 30.5% | $42.52 | 1h 28m | 73.5 | $20.00 | |
| 6 | GPT-6 SolOpenAI | 30.5% | $20.65 | 1h 03m | 76.0 | $10.00 | |
| 7 | Claude Sonnet 5.5Anthropic | 30.1% | $26.52 | 1h 55m | 87.3 | $10.00 | |
| 8 | Claude Fable 5.1Anthropic | 22.9% | $32.23 | 2h 03m | 84.2 | $50.00 | |
| 9 | Claude Opus 5Anthropic | 12.2% | $23.63 | 1h 21m | 83.0 | $25.00 | |
| 10 | Gemini 3.8 FlashGoogle | 11.1% | $27.46 | 2h 28m | 61.6 | $3.75 | |
| 11 | GPT-5.6 TerraOpenAI | 9.5% | $30.86 | 3h 17m | 63.9 | $12.00 | |
| 12 | GLM 5.3Z.ai | 9.2% | $27.47 | 2h 48m | 70.6 | $4.40 | |
| 13 | Claude Haiku 5.5Anthropic | 5.7% | $3.14 | 57m 56s | 70.7 | $0.50 | |
| 14 | Gemini 3.7 FlashGoogle | 4.6% | $21.47 | 1h 50m | 58.0 | $3.75 | |
| 15 | MiMo-V2.6-FlashXiaomi | 4.2% | $0.40 | 3h 22m | 54.7 | $0.28 | |
| 16 | GPT-5.5OpenAI | 3.8% | $17.77 | 46m 28s | 71.3 | $30.00 | |
| 17 | Grok 4.7xAI | 3.8% | $64.77 | 1h 41m | 66.7 | $6.00 | |
| 18 | MiMo-V2.6-ProXiaomi | 3.0% | $0.53 | 2h 23m | 65.4 | $0.87 | |
| 19 | GPT-6 LunaOpenAI | 2.7% | $1.55 | 1h 21m | 57.3 | $0.50 | |
| 20 | Hy4 previewTencent | 2.3% | $4.90 | 1h 40m | 44.3 | $2.501 | |
| 21 | DeepSeek V4 Pro 0813DeepSeek | 1.9% | $5.58 | 2h 20m | 56.1 | $1.98 | |
| 22 | GPT-5.6 LunaOpenAI | 1.5% | $3.44 | 1h 42m | 58.0 | $1.20 | |
| 23 | DeepSeek V4.1 FlashDeepSeek | 0.8% | $0.55 | 1h 01m | 62.5 | $0.84 | |
| 24 | Grok 4.5xAI | 0.8% | $13.42 | 1h 06m | 58.6 | $6.00 | |
| 25 | GLM 5.3 FlashZ.ai | 0.8% | $2.29 | 3h 15m | 57.7 | $0.50 | |
| 26 | Claude Sonnet 5Anthropic | 0.4% | $18.49 | 3h 53m | 57.7 | $10.00 | |
| 27 | Step 5 PreviewStepFun | 0.4% | $3.07 | 1h 34m | 61.1 | $2.70 | |
| 28 | Gemini 3.1 Pro PreviewGoogle | 0.0% | $11.34 | 1h 03m | 32.3 | $12.00 | |
| 29 | Muse Spark 1.2Meta | 0.0% | $23.43 | 1h 40m | 55.7 | $4.25 | |
| 30 | T InklingThinking Machines | 0.0% | $2.23 | 28m 39s | 31.5 | $4.05 | |
| 31 | GLM 5.2Z.ai | 0.0% | $26.58 | 2h 29m | 61.1 | $4.40 |
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.