MARCA (Portuguese, Orchestrator): leaderboard
MARCA (2026) tests LLM web search on 52 manually authored multi-entity questions asked in Portuguese, in the Orchestrator framework (the model delegates sub-questions to subagents that call web_search 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, Orchestrator framework (the model delegates sub-questions to subagents that call web_search and web_scrape); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 88.9 | #145 |
| 2 | GPT-5.2 | 87.2 | #185 |
| 3 | GPT-4.1 | 82.5 | #505 |
| 4 | Gemini 3 Flash | 82 | #175 |
| 5 | GPT-5 Mini | 81.9 | #364 |
| 6 | Sabia-4 | 80 | #440 |
| 7 | Kimi K2 | 77.3 | #443 |
| 8 | Sabiazinho-4 | 75.3 | #668 |
| 9 | GPT-4.1 Mini | 73.5 | #789 |
| 10 | Qwen 3 235B A22B | 67.8 | #628 |
| 11 | Gemini 2.5 Pro | 59.8 | #276 |
| 12 | Gemini 2.5 Flash | 51.7 | #463 |
| 13 | Qwen 3 30B A3B | 49.3 | #1156 |
| 14 | Gemini 2.5 Flash Lite | 35.4 | #964 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=marca-portuguese-orchestrator · How It Works · Data refreshed daily, snapshot 2026-11-21.