Best AI for Portuguese: Claude Opus tied first of 17 LLMs with GPT Sol, GPT Astra and Gemini Pro, at 4 of 5

Which AI is best for Portuguese?

Four of the five Portuguese questions in our Multilingual LLM Bench, run on 2 October 2026, are about Portugal. One is about Brazil. No model of 17 knew the Portugal local fact, and the four leaders got every other question right. So their tie at 4 of 5 is a real tie. Here's how many models got each question:

The one public Portuguese question is a grammar question from Portugal's literary register. It asks for contrafazer, to forge, in the vós conditional that Priberam lists as contrafaríeis, with the pronoun las placed inside the verb. The answer is contrafá-las-íeis, and 12 of the 17 models wrote it. Qwen3.8 Max wrote contrafazê-las-íeis, and Kimi K3 wrote contrafar-las-íeis. You can read every answer on our public question page.

The other four stay private, so we can rerun them on new models. The Brazil question split the field. Only 6 of 17 got it, and all four leaders were among them.

This page ranks the chatbots that write, translate and answer in Portuguese. If you want to learn Portuguese, that's a different pick: tutor apps such as Langua and Speak are built around speaking practice, and none of them was in our test.

How did 17 models do on Whizi's own Portuguese questions?

Four models tied first with 4 of 5, and two more got 3. Every model got the same 5 questions in each of 30 languages, at low reasoning. We graded each answer against a sourced key, and an answer in the wrong language scored zero.

Five questions is a small sample. One question moves a model's Portuguese score by 20 points, so read this as a rough order, not a fine ranking.

Portuguese split the field harder than French did. Four of the models got none of the five right, and two of those were from Qwen. One of them, DeepSeek V4.1 Flash, sent back an empty answer to the public grammar question, so it had no chance there.

Claude Fable cost more to run than Claude Opus, $0.66 against $0.47, and got half its score. GPT Luna, the cheapest model we ran at $0.01 for all 150 questions, got 3. It was one of the six that knew the Brazil answer.

Don't use our bench to rank Portuguese against other languages. Portuguese averaged 35.3% across the 17 models, against 51.8% for French. That says more about how hard each set of five questions was than about the language.

What do the public Portuguese leaderboards say?

The public leaderboards put Claude Opus and Gemini Pro at the top in Portuguese, but only by one point. Artificial Analysis scores Portuguese on its Multilingual Index. As we read it on 25 October 2026, Gemini 3.1 Pro Preview and Claude Opus 4.6 (max) tie for first at 94, with Gemini 3 Pro Preview (high), Claude Opus 4.6 (non-reasoning, high) and Grok 4 at 93. The page doesn't say if it tests Brazilian or European Portuguese. Claude Opus 5.5, the GPT-6 models and Gemini 4 weren't on it when we looked.

Both of the vendors test Brazil. Anthropic's multilingual support page says that Claude Sonnet 4.5 keeps 97.8% of its English score in Portuguese (Brazil), and that Claude Haiku 4.5 keeps 96.1% of it. OpenAI publishes its own numbers in its simple-evals results, on an MMLU test that professional translators rewrote. There, its o3 model scores 0.910 in Brazilian Portuguese. GPT-4.1 nano, the lowest scorer on that page, got 0.741.

Neither vendor table has a European Portuguese row. Our questions lean the other way, so the two kinds of evidence cover each other's gap. The AMALIA team's technical report names the same gap: it says European Portuguese is underrepresented in both training data and native evaluation.

Does AI write Brazilian or European Portuguese?

AI writes the Portuguese your prompt asks for, so say Brazil or Portugal. Words are the first split. The Priberam dictionary, which marks both norms, gives autocarro as the Portugal word for a bus, with ônibus as the equivalent in Brazil. Its entry for telemóvel does the same with celular.

Spelling is the second split, and it's still there after the 1990 spelling agreement. The Acordo Ortográfico de 1990 keeps double spellings where the educated pronunciation differs: "facto e fato", "recepção e receção". It keeps the accent split too, in pairs like género and gênero or António and Antônio. A text that mixes the two of them reads as neither.

Address is the third. Pick você, tu or o senhor before you start, and tell the model, because a draft that drifts between them reads as careless in both countries.

```

Write in Portuguese for readers in [Brazil / Portugal].

Follow the [Brazilian / European] spelling of the 1990 agreement, for example [fato / facto] and [gênero / género].

Use [você / tu / o senhor] for the reader, and keep it the same all the way through.

Use the everyday words people in that country use, for example [ônibus / autocarro] and [celular / telemóvel].

Keep every name, number, date and amount exactly as written.

At the end, list any word or phrase that would sound foreign in that country.

Task: [a product email / a school essay / a WhatsApp message] about [topic], about [200] words.

```

For translation, swap the task line for "Translate this English text into Portuguese for [Brazil / Portugal]" and keep the rest. Our guide to using AI to translate covers the checks to run after.

Should you use AMALIA, Sabiá or LatamGPT?

Use AMALIA, Sabiá or LatamGPT when you want a model built for Portuguese, or one you run yourself. None of the three was in our run or in Artificial Analysis's Portuguese top five, so we can't rank them against Claude Opus or GPT.

AMALIA is Portugal's own model. Its model card says it's for European Portuguese and that the project "is funded by the Government of Portugal's Development and Innovation Programmes". The tech report came out on 27 March 2026. It says "AMALIA matches strong baselines on translated benchmarks" and does better on tests written for Portugal.

Sabiá is the model you pay for in reais. Maritaca's API page calls Sabiá 4 Thinking "Our most advanced model, with reasoning", and lists Sabiazinho 4 at R$1.00 in and R$4.00 out per million tokens as "Our fastest model".

LatamGPT is infrastructure, not a chatbot. Its model card lists "Spanish, Portuguese, and English, with a special focus on Latin American variants". It doesn't list any benchmark results.

Qwen3-8B is the small general model to try if you'd rather not run a Portuguese specialist. Its model card claims "Support of 100+ languages and dialects", but it gives no Portuguese score. We didn't run it. Qwen's two hosted models in our run, Qwen3.8 Max and Qwen3.8 Flash, both got 0 of 5.

How much do ChatGPT, Claude and Gemini cost in Brazil and Portugal?

Brazil charges 30% more for Claude Pro than it does for ChatGPT Plus: it's R$129.90 against R$99.90 a month, or $26.13 against $20.10. We read each app's App Store listing in Brazil and Portugal on 25 October 2026. The dollar figures use the Currency API rates of 24 October 2026, rounded to the cent.

Portugal pays the same euro prices as France for all six plans. In both countries, Google AI Pro is the cheapest of the three main plans. Claude Pro is the dearest in Brazil and the middle one in Portugal.

If you want GPT Astra, one of the four leaders, you won't get it in ChatGPT Plus chats. OpenAI's help page says "GPT-6 Pro, powered by GPT-6 Astra, is available in ChatGPT for Pro $100, Pro $200, Business, and Enterprise plans." Plus gets GPT Sol in chat, and that's the model that tied Astra in our Portuguese test. Claude Pro does include Claude Opus, as claude.com/pricing shows. For more countries, see Claude Pro price by country. If your card keeps failing, paying for AI when your card is declined covers what you can do about it.

Where does Whizi fit for Portuguese?

Whizi is our product, so here's where it helps and where it loses. Pro at $29.99 has three of the four models that tied first in our Portuguese test, Claude Opus, ChatGPT Sol and Gemini Pro, with side by side model chats to compare them on one prompt. GPT Astra, the fourth, is on Powerhouse at $49.99, and since credits per message vary by model and Astra is one of the dearest, it adds little for Portuguese that Pro doesn't already give you. Whizi loses on a single plan: Claude Pro alone costs less in both stores, and ChatGPT Plus is R$99.90 ($20.10) in Brazil. AMALIA, Sabiá and LatamGPT aren't in Whizi's catalogue. Side by side model chats shows how compare works.