MLOps-Bench - Base ReAct - Test Pass: leaderboard
MLOps-Bench (2026), practical ML engineering tasks in reproducible environments; Base ReAct harness, published evaluation split with separate development tasks. Task completion requires every task test to pass; test pass averages partial correctness. The three agent harnesses are separate boards; effort and API snapshots not supplied by the paper are not inferred. Rate (%); higher is better.
Metric: Rate (%). Source: arxiv.org. Saturation forecast: Around August 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 47.9 | #31 |
| 2 | Claude Sonnet 4.6 | 39.2 | #90 |
| 3 | GPT-5.2 | 31.4 | #110 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mlops-bench-base-react-test-pass · How It Works · Data refreshed daily, snapshot 2026-10-20.