TensorBench - Codegen: leaderboard
TensorBench (2026) tests coding agents (a published CLI harness with its model) on feature-addition and refactoring tasks in Scorch, a compiler-based tensor framework with dense and sparse tensors; a task passes when the patched repository passes its randomized regression tests and the tests the agent adds. This board is the pass rate in percent on the 7 code-generation tasks, such as OpenMP parallelism, SIMD vectorization and a CUDA backend; higher is better.
Metric: Pass rate (%) on the 7 code-generation tasks (OpenMP and atomic-scatter parallelism, SIMD vectorization, CUDA and out-of-core backends); 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 February 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 (xHigh) | 71.4 | #79 (Claude Opus 4.7) |
| 2 | Claude Opus 4.6 | 57.1 | #102 |
| 3 | GPT-5.5 (xHigh) | 57.1 | #49 (GPT-5.5) |
| 4 | GPT-5.3 Codex (xHigh) | 42.9 | #116 (GPT-5.3 Codex) |
| 5 | Gemini 3.1 Pro (Preview) | 28.6 | #94 |
| 6 | Qwen 3 Coder 480B A35B Instruct | 14.3 | #643 |
| 7 | GPT-5.4 (xHigh) | 14.3 | #123 (GPT-5.4) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=tensorbench-codegen · How It Works · Data refreshed daily, snapshot 2026-11-13.