SWE-ContextBench (Oracle Summary): leaderboard
Metric: Resolved rate (%): share of the 376 related tasks whose patch passes every fail-to-pass and pass-to-pass test (SWE-ContextBench: 376 related GitHub issues across 39 repositories and 8 languages, linked by issue and pull request references to 1,100 SWE-Bench Lite, Multilingual and Verified experience tasks; each model runs inside its coding-agent harness in a fresh repository checkout; oracle summary setting: the agent is given the gold related experience task's summary (about 217 tokens, the final summary of a Claude Code with Claude Sonnet 4.5 trajectory)); higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.3 Codex | 23.94 | #68 |
| 2 | Claude Sonnet 4.5 | 23.4 | #138 |
| 3 | MiniMax-M2.7 | 22.07 | #248 |
| 4 | Qwen 3.5 Plus | 21.54 | #123 |
| 5 | GLM-5 | 19.41 | #137 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=swe-contextbench-oracle-summary · How It Works · Data refreshed daily, snapshot 2026-10-11.