🇵🇹 Best AI for Portuguese
Use this one
- Best overall: zai-org/GLM-5.3-Flash. Statistically close: gpt-6-astra, gemini/gemini-3.7-flash, gpt-6.1-sol.
- Best open-weight (self-host / cheap): zai-org/GLM-5.3-Flash (rank score 1.19).
- Best for writing (summaries + grammar, top 40 models): zai-org/GLM-5.3-Flash (56.58).
- Translation & agent tasks: no per-model Portuguese benchmark data yet in EuroEval's public generative board, so we don't rank them. See gaps.
| # | Model | Rank score ↓ | Writing ↑ | Understanding ↑ |
|---|---|---|---|---|
| 1 | zai-org/GLM-5.3-Flash open | 1.19 | 56.58 | 80.45 |
| 2 | gpt-6-astra | 1.23 | 54.65 | 82.88 |
| 3 | gemini/gemini-3.7-flash | 1.24 | 53.0 | 82.64 |
| 4 | gpt-6.1-sol | 1.24 | 54.82 | 83.55 |
| 5 | gemini/gemini-3.6-flash | 1.33 | 52.78 | 82.01 |
| 6 | deepseek/deepseek-flash#no-thinking | 1.35 | 53.34 | 74.95 |
| 7 | zai-org/GLM-5.3-Flash open | 1.40 | 55.14 | 77.04 |
| 8 | Qwen/Qwen3.6-27B-FP8 open | 1.40 | 49.17 | 74.25 |
| 9 | Qwen/Qwen3.8-27B open | 1.41 | 49.14 | 74.27 |
| 10 | gpt-5.6-sol | 1.45 | 49.69 | 80.51 |
| 11 | gpt-6-sol | 1.46 | 50.91 | 78.92 |
| 12 | google/gemma-4-31B-it open | 1.50 | 51.05 | 71.26 |
| 13 | claude-sonnet-4-6 | 1.53 | 54.88 | 70.36 |
| 14 | gemini/gemini-3-flash-preview#thinking | 1.54 | 50.99 | 81.06 |
| 15 | Qwen/Qwen3.5-35B-A3B-FP8 open | 1.58 | 46.7 | 70.62 |
Source: EuroEval Portuguese generative leaderboard, retrieved 2026-10-15. 185 models evaluated. Datasets: alba_mcq_pt, cultura_viva_pt, harem, multi_wiki_qa_pt, publico, scala_pt, sst2_pt, winogrande_pt.
The smartest model isn't always deployable.
Get ahead of the constraints with The Signal.