DeepWeb-Bench - Retrieval: leaderboard

The retrieval capability family of DeepWeb-Bench (2026), a deep-research benchmark that demands massive cross-source web evidence and long-horizon derivation, scored cell by cell. The score is the mean cell score (%) over the retrieval cells for agents run in Claude Code or Codex CLI with the search and browsing tools the benchmark provides; higher is better.

Metric: Mean cell score (%) over the Retrieval capability family (finding the evidence cells), DeepWeb-Bench deep-research tasks answered cell by cell with the benchmark's web search, page visit and PDF tools (native browsing disabled), runs hosted in Claude Code CLI or Codex CLI at default settings; higher is better. Source: arxiv.org. Saturation forecast: Around July 2028. 9 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5.537.84#56
2Claude Opus 4.736.52#86
3GLM-5.134.19#200
4Claude Sonnet 4.633.8#179
5DeepSeek V4 Flash33.72#273
6DeepSeek V4 Pro32.89#201
7Qwen 3.6 Plus32.25#245
8MiniMax-M2.728.56#539
9Kimi K2.626.21#197

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=deepweb-bench-retrieval · How It Works · Data refreshed daily, snapshot 2026-11-22.