InterLV-Search - Level 1: leaderboard
Level 1 (active visual evidence seeking over an offline multimodal corpus, at most three interactions) of InterLV-Search (2026), an interleaved language-vision agentic search benchmark in which intermediate visual evidence decides what to search next; agents use image search, reverse image search, web search, browsing, cropping and code tools. Final-answer accuracy judged by GPT-5.4-mini, percent; higher is better.
Metric: Level 1 (active visual evidence seeking, 975 questions over an offline multimodal corpus, at most 3 interactions) final-answer accuracy judged by GPT-5.4-mini for semantic equivalence, %, through the InterLV-Agent reason-act-observe framework with image search, reverse image search, web search, browsing, cropping and code tools; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 46.05 | #99 |
| 2 | GPT-5.4 | 45.64 | #135 |
| 3 | GPT-5 | 39.54 | #178 |
| 4 | Claude Sonnet 4.6 | 37.13 | #173 |
| 5 | Qwen 3.6 Plus | 29.27 | #238 |
| 6 | SenseNova-MARS-32B | 15.69 | |
| 7 | MMSearch-R1-7B | 4.1 | |
| 8 | VDR-8B | 3.13 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=interlv-search-level-1 · How It Works · Data refreshed daily, snapshot 2026-11-21.