Coding benchmark
SWE-bench (Vals) leaderboard
SWE-bench, Vals AI run. Every model the catalog carries a published SWE-bench (Vals) value for, ranked by that value.
Vals AI’s independent run of the public SWE-bench issue set, reported by human time-to-fix bucket. Vals removed SWE-bench Verified from its index as saturated on 2026-05-04; this board is the standalone SWE-bench run.
SWE-bench (Vals) ranking
50 models with a published SWE-bench (Vals) value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Resolved rate |
|---|---|---|---|
| 1 |  Claude Opus 5 | Anthropic | 97.0 |
| 2 |  DeepSeek V4 Pro 0813 | DeepSeek | 96.4 |
| 3 |  GPT-5.6 Sol | OpenAI | 96.2 |
| 4 |  Grok 4.6 | xAI | 95.6 |
| 5 |  GLM-5.3 | Z.AI | 95.4 |
| 5 | GPT-5.6 Terra | OpenAI | 95.4 |
| 7 | Claude Fable 5 | Anthropic | 95.0 |
| 8 |  Kimi K3 | Moonshot AI | 93.4 |
| 9 | GPT-5.6 Luna | OpenAI | 93.0 |
| 10 | GLM-5.3-Flash | Z.AI | 92.0 |
| 11 | DeepSeek V4 Flash 0731 | DeepSeek | 88.8 |
| 12 | Claude Opus 4.8 | Anthropic | 88.6 |
| 13 | Grok 4.5 | xAI | 86.6 |
| 13 |  Muse Spark 1.2 | Meta | 86.6 |
| 15 |  Qwen3.8-27B | Alibaba | 86.0 |
| 16 | Qwen3.8 Max | Alibaba | 85.6 |
| 17 | GLM-5.2 | Z.AI | 82.8 |
| 18 | GPT-5.5 | OpenAI | 82.6 |
| 19 | TMInkling-Small | Thinking Machines Lab | 82.2 |
| 20 | Claude Opus 4.7 | Anthropic | 82.0 |
| 20 | Muse Spark 1.1 | Meta | 82.0 |
| 22 |  Gemini 3.7 Flash | 80.8 | |
| 23 | Gemini 3.8 Flash | 80.0 | |
| 24 | Claude Sonnet 5 | Anthropic | 79.6 |
| 24 | Gemini 3.6 Flash | 79.6 | |
| 26 | Gemini 3.1 Pro | 78.8 | |
| 26 | Gemini 3.5 Flash | 78.8 | |
| 28 | TMInkling | Thinking Machines Lab | 77.6 |
| 29 | Claude Sonnet 4.6 | Anthropic | 77.4 |
| 30 | GLM-5.1 | Z.AI | 76.4 |
| 31 | Kimi K2.6 | Moonshot AI | 76.2 |
| 32 | Gemini 3 Flash | 75.0 | |
| 32 | Gemini 3.5 Flash-Lite | 75.0 | |
| 32 |  MiniMax M3 | MiniMax | 75.0 |
| 35 |  MiMo-V2.5-Pro | Xiaomi | 74.0 |
| 36 | MiniMax M2.7 | MiniMax | 73.8 |
| 37 | Qwen3.6 Plus | Alibaba | 73.4 |
| 38 | GPT-5.4 mini | OpenAI | 73.0 |
| 39 | Grok 4.20 | xAI | 72.2 |
| 40 | Grok 4.3 | xAI | 71.4 |
| 41 | MiMo-V2.5 | Xiaomi | 71.0 |
| 42 | GPT-5.4 nano | OpenAI | 69.8 |
| 43 |  Nemotron 3 Ultra | NVIDIA | 69.0 |
| 44 | Qwen3.7 Max | Alibaba | 68.8 |
| 45 | Claude Haiku 4.5 | Anthropic | 66.6 |
| 46 |  Mistral Medium 3.5 128B | Mistral | 66.4 |
| 47 |  Ling 3.0 Flash | InclusionAI | 65.2 |
| 48 | Gemini 3.1 Flash-Lite | 62.8 | |
| 49 |  Laguna M.1 | Poolside | 57.6 |
| 50 | Laguna XS.2 | Poolside | 55.2 |
Evidence key: Observed
Rows are ordered by the value Vals AI SWE-bench, Vals AI run leaderboard published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.