Best AI for Italian: Gemini Pro leads from €21.99 a month, and 16 of 17 models missed one Italian saying

Which AI is best for Italian?

Gemini Pro is the best AI for Italian, and GPT and Claude Opus are level with it on our own questions. Gemini is at the top of the Artificial Analysis Italian index, and it tied for first on the one recent Italian translation test that was graded by people. When we ran 17 models, five of them tied on 4 of 5 Italian questions, and none of them got all five. Most of the misses came from two of the questions:

Sixteen of the 17 models missed the second saying, and Claude Opus was the only one that could name it. Only four models got the second local fact, and all four of them were from Google and OpenAI. So no model got both of our two hardest questions.

How did 17 models do on Whizi's Italian questions?

Five models tied for first, and each of them got 4 of the 5 Italian questions right: two from OpenAI, two from Google and one from Anthropic. That's from our Multilingual LLM Bench, run on 2 October 2026. 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 answer moves a score by 20 points, so treat it as a rough order.

The public question is the grammar one. It gives the model an Italian sentence that means "I don't think this matters to him" and asks it to swap in the old literary verb calere, in the present subjunctive, with both of the objects turned into pronouns. Treccani has the verb, and the answer is "gliene caglia". Thirteen of 17 models wrote it.

The four misses were all near ones, and all four came from the cheaper models: Gemini Flash-Lite, Claude Sonnet, GLM Flash and Qwen Flash. Each of them wrote a form that is one step off, such as "gliene cagli". That's the kind of slip a fluent reader would catch and a learner wouldn't.

Don't use our bench to rank Italian against other languages. The 17 models averaged 50.6% in Italian and 61.2% in English, and that tells you more about how hard each set of five questions was.

Is ChatGPT good at Italian?

Yes, ChatGPT is good at Italian. OpenAI's simple-evals results put its o3 reasoning model at 0.912 in Italian, on an MMLU test that professional translators rewrote in 14 languages. GPT-4.1 scored 0.869 and the small GPT-4.1 nano 0.734. In our test, GPT Sol and GPT Astra both got 4 of 5.

Which GPT you get depends on the plan. OpenAI's help page on GPT-6 says Free and Go now default to GPT-5.6 Luna, Plus users get GPT-6 Sol, and Astra powers Chat only on Pro.

Claude keeps nearly all of its English skill when it works in Italian. Anthropic's multilingual support page says that Claude Sonnet 4.5 kept 97.9% of its English score in Italian, and that of the 14 languages it lists, only Spanish did better, at 98.2%. The smaller Claude Haiku 4.5 kept 96.0%.

Which AI is best for Italian translation?

Gemini Pro and GPT are the best big chatbots for English to Italian, tied in the top group of the one recent test graded by people. The WMT 2025 findings had native Italian speakers score the shortlisted systems from 0 to 100. Here's the English to Italian table:

Italian was one of the two hardest pairs. The organisers wrote that "English → Arabic (Egyptian) and English → Italian proved to be the most challenging." Claude 4 placed eighth here, 7.3 points behind Gemini. The graders were in house annotators at Cohere, the maker of the Command A models in the table, and the models are a year old: no GPT-6, Claude Opus 5 or Gemini 3 model was in it.

For a plain translation, DeepL gives you one thing a chatbot doesn't: a switch for Lei or tu. Its formality feature lists Italian among its languages, for DeepL Pro and API subscribers. With a chatbot, write the register into the prompt:

```

Traduci questo testo in italiano, per [clienti / colleghi / un ente pubblico].

Dai del [Lei / tu] al lettore in tutto il testo.

Mantieni esattamente nomi, numeri, date e importi.

Scrivi i numeri all'italiana: 1.234,56 € e 25/10/2026.

Usa le virgolette basse «».

Se un modo di dire non ha un equivalente italiano, traducine il senso.

Alla fine elenca le parole su cui hai avuto dubbi.

[incolla qui il testo]

```

Our guide to using AI to translate has the checks to run after.

Can you use DeepSeek in Italy?

No, not through the app stores. Italy's data protection authority, the Garante, ordered DeepSeek to stop processing the data of people in Italy on 30 January 2025. The order took effect on the day DeepSeek received it, and it was the regulator's answer to DeepSeek's claim that it had never entered the Italian market. On 25 October 2026, Apple's Italian App Store still returned no page for the DeepSeek app, while the US store listed version 2.7.1.

Italy has done this before. The same authority ordered a temporary limit on ChatGPT on 30 March 2023, and ChatGPT is back on sale in Italy today. DeepSeek isn't, so for Italian the choice is Gemini, ChatGPT, Claude or an Italian model. Our page on DeepSeek bans by country shows that Italy is the one country where ordinary users still can't get the app.

Is there an Italian AI model?

Yes, two of them are worth knowing: ChatMinerva from Sapienza University of Rome and Velvet from Almawave. Neither is near the top of the Italian leaderboards, so pick one for where it was built, not for raw quality. The one model in the Artificial Analysis Italian top five that isn't from Google or Anthropic is Alibaba's Qwen3.5 397B, at 93, and it isn't Italian either.

Avvenire reported the ChatMinerva launch on 3 June 2026. Roberto Navigli, who leads the Sapienza team, put the case for it this way: "la lingua sono gli occhiali con cui l'AI vede il mondo" (language is the glasses through which AI sees the world). Avvenire gave no score for it.

Almawave grades Velvet itself. Its model card reports 58.6 on Italian MMLU. OpenAI reports 0.912 for o3 on its own Italian MMLU, a different setup, but the gap is wide. That's why Velvet makes sense on your own servers, if you run a company that has to keep its data in house. It isn't the one to pick as your daily chatbot.

How much do ChatGPT, Claude and Gemini cost in Italy?

ChatGPT Plus costs €22.99 a month in Italy's App Store, Claude Pro €22.00 and Google AI Pro €21.99, read on 25 October 2026. Google AI Pro is the cheapest plan with the model that leads the Italian index:

Google AI Plus is the cheap surprise. At €4.99 it gives you Google's Flash model, and a Gemini Flash model tied first in our Italian test. Google's own Italian plans page lists the same €4.99 and €21.99, and adds Gemini 4 Argon to Pro. That model is newer than our run, so we haven't tested its Italian. For other countries, see ChatGPT Plus prices by country.

Which AI is best for learning Italian?

Claude Opus or GPT is the best AI for learning Italian, because both tied first on our questions and Opus was the only model to name our second saying. Ask it to correct you and to name the rule each mistake breaks:

```

Sei il mio insegnante di italiano. Il mio livello è A2.

Parlami in italiano semplice e dammi del tu.

Dopo ogni mia risposta, correggi i miei errori in inglese e spiega la regola (congiuntivo, articoli, preposizioni articolate).

Poi continua la conversazione.

Argomento: ordinare al bar.

```

For speaking practice, Duolingo's blog said on 14 January 2026 that its AI Video Call covers Max subscribers learning Italian on iOS. Check any proverb a chatbot gives you against a dictionary, since 9 of 17 models missed our first one.

Where does Whizi fit for Italian?

Whizi is our product, so here's where it helps and where it loses. Pro at $29.99 a month includes Gemini Pro, GPT Sol, Claude Opus and Mistral Large, and its side by side model chats run one Italian prompt through two of them at once. Buyers in Italy are charged in euros at checkout. GPT Astra needs Powerhouse at $49.99. Whizi loses on two counts. It doesn't have DeepL, and it has no EU data region: its privacy policy names Canada and the United States as places where processing may happen.