Prosa: leaderboard
Rankings produced by holistic LLM-as-a-judge scoring are sensitive to the bias of the chosen judge model. We show that switching to binary rubric scoring with multi-judge filtering removes this sensitivity: decomposing the judgement matters more than the judge model itself. To support this claim, we introduce Prosa, the first real user multi-turn Brazilian Portuguese chat benchmark: 1,000 WildChat
Metric: Score (self-reported). Source: benchmarklist.com. Saturation forecast: Around March 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 88.6 | #110 |
| 2 | GPT-5 Mini | 84.8 | #186 |
| 3 | Gemini 3 Pro | 83.8 | #80 |
| 4 | Qwen 3 235B A22B | 80.1 | #313 |
| 5 | Gemini 3 Flash | 78.8 | #99 |
| 6 | Gemini 2.5 Flash | 76.3 | #245 |
| 7 | GPT-4.1 | 74.8 | #248 |
| 8 | Qwen 3 30B A3B | 73.9 | #493 |
| 9 | GPT-4.1 Mini | 67.7 | #355 |
| 10 | Qwen 3 4B | 64.6 | #829 |
| 11 | GPT-4o | 55 | #343 |
| 12 | GPT-4o Mini | 54.5 | #599 |
| 13 | Qwen 2.5 14B | 46.4 | #725 |
| 14 | Qwen 2.5 7B | 43.3 | #922 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=prosa · How It Works · Data refreshed daily, snapshot 2026-10-20.