MERRIN (Native Search): leaderboard

MERRIN (2026) asks 162 human-written questions whose single short answers depend on non-text web evidence (images, video, audio or tables) found among noisy and conflicting sources, with no hint of the modality in the question. Native setting with the built-in search and page-reading tools of each vendor: accuracy over three runs, judged against the gold answer, in percent; higher is better.

Metric: Accuracy (%) with the vendor's built-in search (GPT web_search; Gemini Google Search plus URL Context; the Gemini Deep Research agent), on MERRIN's 162 human-annotated questions with unique short answers that need non-text web evidence (image, video, audio or table) and multi-hop reasoning or conflict resolution across noisy sources; accuracy against the gold answer by an LLM judge with the BrowseComp prompt, mean of three runs; higher is better. Source: arxiv.org. Saturation forecast: Around August 2027. 7 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini Deep Research Agent33.3
2Gemini 3 Flash31.7#175
3Gemini 3.1 Pro (Preview)29#99
4Gemini 3 Pro28.8#145
5Gemini 3.1 Flash Lite20.6#371
6GPT-5.4 Mini15.6#406
7GPT-5.4 Nano12.6#651

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=merrin-native-search · How It Works · Data refreshed daily, snapshot 2026-11-21.