MARCA (Portuguese, Basic): leaderboard

MARCA (2026) tests LLM web search on 52 manually authored multi-entity questions asked in Portuguese, in the Basic framework (the model itself calls web_search via the Serper API and web_scrape). A GPT-4.1 judge checks each answer against a manually validated checklist; checklist accuracy (%) averaged over three runs, higher is better.

Metric: Checklist accuracy (0-1, times 100): share of a question's checklist items (expected entities and attributes) that a GPT-4.1 judge marks satisfied, averaged over MARCA's 52 Portuguese questions and three runs, Basic framework (the model itself calls web_search via the Serper API and web_scrape); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro89.6#146
2GPT-5.285.5#190
3GPT-5 Mini85.2#373
4GPT-4.182.6#506
5Kimi K279.7#448
6Gemini 3 Flash73.3#180
7Sabiazinho-470.5#672
8Sabia-469.5#445
9Qwen 3 235B A22B67.7#633
10GPT-4.1 Mini63.1#799
11Gemini 2.5 Pro58.5#280
12Qwen 3 30B A3B48.4#1165
13Gemini 2.5 Flash47.5#466
14Gemini 2.5 Flash Lite29.9#972

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

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