Search-SWE - Task - Long-PDF Sentence Evidence — Vectorless: leaderboard
Search-SWE Long-PDF Sentence Evidence — Vectorless: coding agents build working retrieval systems under executable, resource and compliance checks. The continuous task score measures retrieval quality rather than a binary solve rate; overall and hardware subsets average tasks equally. The source-selected Codex v0.164.0 runs retain model effort and exact task/verifier configuration, including graded valid timeout submissions.
Metric: Mean retrieval task score (%). Source: search-swe.github.io. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4.1 Flash 1017 (Max) (Search-SWE Run: Codex v0.164.0) | 50 | |
| 2 | GPT-6 Luna (Max) (Search-SWE Run: Codex v0.164.0) | 8.33 | |
| 3 | MiMo-V2.6-Flash (High) (Search-SWE Run: Codex v0.164.0) | 0 |
Interactive version: theaggregate.ai/benchmark?slug=search-swe-task-long-pdf-sentence-evidence-vectorless · How It Works · Data refreshed daily, snapshot 2026-11-05.