Coding benchmark
FrontierSWE v2 leaderboard
Every model the catalog carries a published FrontierSWE v2 value for, ranked by that value.
CategoryCoding
MeasureFive-trial mean task score (Mean@5), 0-100
Tasks34 ultra-long-horizon engineering and research tasks
DifficultyUltra-long-horizon frontier software engineering
Published byFrontierSWE v2
A 34-task expansion of FrontierSWE for ultra-long-horizon engineering and research work that remains far from saturation.
FrontierSWE v2 ranking
12 models with a published FrontierSWE v2 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 | Five-trial mean task score (Mean@5), 0-100 |
|---|---|---|---|
| 1 |  Claude Fable 5.1 | Anthropic | 56.3 |
| 2 | Claude Opus 5 | Anthropic | 52.0 |
| 3 | Claude Fable 5 | Anthropic | 47.0 |
| 4 |  GPT-5.6 Sol | OpenAI | 32.2 |
| 5 |  GLM-5.3 | Z.AI | 30.2 |
| 6 |  Kimi K3 | Moonshot AI | 25.9 |
| 7 |  Grok 4.6 | xAI | 25.3 |
| 8 |  Gemini 3.7 Flash | 20.3 | |
| 9 | Gemini 3.8 Flash | 19.6 | |
| 10 |  Qwen3.8 Max | Alibaba | 15.8 |
| 11 |  Muse Spark 1.2 | Meta | 12.0 |
| 12 | TMInkling | Thinking Machines Lab | 4.1 |
Evidence key: Observed
Rows are ordered by the value FrontierSWE v2 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.