Maritaca AI

LLMs specialized for Portuguese

[Maritaca AI @MaritacaAI]

S5, our newest model, is now available through the API — and it's free to try for a limited time!

To help you explore what it can do, we're offering S5-free with no token charges. Just select "S5-free" as the model in your API requests.

This is a chance to put S5 to work on your own use cases and see how it performs in your application's day-to-day tasks. Bring your prompts, test it on your workflows, and explore what you can build with it.

The free version accepts text only. Built-in tools and PDF and image input are available through the paid API.

Get started with the documentation in English: https://t.co/duzy1EA1j5

[Maritaca AI @MaritacaAI]

[OPEN WEIGHTS]

Meet S5-nano, Maritaca's new open-weight model. Anyone — individuals, companies or universities — can download it, run it on their own infrastructure, and adapt it to their needs.

S5-nano has 4 billion parameters and is built from Qwen3.5-4B, with continued pretraining and post-training carried out by our team.

To see how it compares with other open-weight models, we evaluated 10 models, including Brazilian projects. S5-nano ranks first on average across native Portuguese benchmarks in both reasoning and non-reasoning modes. The evaluation covers law, medicine, exams and agentic tasks, among other areas.

Made in Brazil by Maritaca, S5-nano gives companies, universities and government organizations a model they can run and adapt themselves, on infrastructure they control.

Download the weights and read the full results and evaluation protocol in English:
https://t.co/pGgdV29WbO

Evaluation repository:
https://t.co/HPM0PTBow8

[Maritaca AI @MaritacaAI]

Introducing S5, Maritaca's most advanced model.

With S5, you can create a game that runs in the browser, build a 3D simulation, or turn hundreds of documents into a presentation. It has more autonomy to plan tasks and use tools, including inside coding agents, and supports up to 1 million tokens of context to work with large amounts of information.

Our launch blog brings together games, simulations and other interactive examples you can try for yourself.

We're keeping S5-free available on the API for a limited time so you can test the model in your own projects without token charges. This version accepts text only and does not include built-in tools or PDF and image input.

Those capabilities are available through the paid API. S5 costs R$0.50 per million input tokens and R$2.50 per million output tokens, with additional charges for tools and PDF and image processing.

We've also temporarily increased the free plan's limits in Maritaca chat so more people can try S5. Chat Premium costs R$30 per month when billed annually.

Explore the demos: https://t.co/GBwK1tgpyy
Try the chat: https://t.co/ZD0T9zupKM
API documentation: https://t.co/duzy1EAz8D

[Maritaca AI @MaritacaAI]

Also available for free to everyone at https://t.co/Uy4ys3YhKC

[Maritaca AI @MaritacaAI]

Attention AI system builders 🤖:

Sabiá-3 is now also available via API!

It's the best LLM in Portuguese, at the best price: 10 USD for every 1 million tokens (input or output)

Instructions to use the API at: https://t.co/L8zdZ8DqBM

[Maritaca AI @MaritacaAI]

@messiashenrique We plan to make it available next Monday

[Maritaca AI @MaritacaAI]

Sabiá-3 is here!

Sabiá-3 is the newest and most advanced language model from Maritaca AI, with accuracy comparable to GPT-4o in 64 Brazilian exams, including OAB, Enem, and various professions from ENADE.

Start using it at https://t.co/P4jPUeo7MH 😍

[Maritaca AI @MaritacaAI]

@marcoz_auguzto Hi @marcoz_auguzto , yes, it should know it

[Maritaca AI @MaritacaAI]

Para mais informações, confira o blog post: https://t.co/PV91rU0aTH
E o relatório técnico: https://t.co/x4Glbn6Vos

[Maritaca AI @MaritacaAI]

Quase um ano atrás, lançamos o Sabiá-65B , o primeiro LLM para o Português, com desempenho ligeiramente superior ao GPT-3.5-turbo.

Hoje, temos o prazer de anunciar o lançamento do Sabiá-2, a mais recente geração de LLMs dedicados ao Português.

[Maritaca AI @MaritacaAI]

No quesito preço vs desempenho, o Sabiá-2 Medium tem o mesmo desempenho do Mistral Large mas é 8x mais barato Teste agora em https://t.co/Uy4ys3XJV4 Os modelos também estão disponíveis via API e em suas versões "baixáveis" via MariTalk Local.