TensorBench: leaderboard
Metric: Pass rate (%) on all 199 feature-addition and refactoring tasks; each coding agent (a published CLI scaffold with its model) edits the open-source Scorch compiler-based tensor framework inside a per-task container, one run per task; a task passes when every test of the post-patch suite (pre-existing randomized regression tests plus the tests the agent adds) succeeds; higher is better. Source: arxiv.org. Saturation forecast: Around August 2027. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.7 (xHigh) | 64.8 |
| 2 | GPT-5.5 (xHigh) | 58.8 |
| 3 | Claude Opus 4.6 | 42.7 |
| 4 | GPT-5.4 (xHigh) | 38.7 |
| 5 | GPT-5.3 Codex (xHigh) | 36.2 |
| 6 | Gemini 3.1 Pro (Preview) | 31.7 |
| 7 | Qwen 3 Coder 480B A35B Instruct | 22.1 |
Interactive version: theaggregate.ai/benchmark?slug=tensorbench · How It Works · Data refreshed daily, snapshot 2026-09-29.