Maritaca AI (Sabiá)
Brazilian Portuguese LLMs beating GPT on Brazil's bar exam and ENEM
Overview
Maritaca AI is a Unicamp-spinout Brazilian lab building the Sabiá LLM family (MariTalk) for Portuguese and Brazilian legal, tax and educational context, sold through an OpenAI-compatible API priced in reais with a free web and mobile chat tier.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 3 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- 128K context
- Input/output priced separately
- Free trial available
- 128K context
- Suited to most production apps
- 128K context
- Sabiá-4 Thinking reasoning variant available
- Priority support
- Volume-based quotes
What you can produce with Maritaca AI (Sabiá)
- Sabiá-4, Sabiá-3.1 and Sabiázinho-4 models via OpenAI-compatible API
- 128,000-token context window across all models
- Free web chat and mobile apps (MariTalk)
- Sabiá-4 Thinking reasoning model with cited step-by-step answers
- Live queries against Brazilian public databases (Central Bank, IBGE, TSE)
- In-country inference option (-br-sp suffix) for data residency
Maritaca started in October 2022, a month before ChatGPT, as a Unicamp research spinout under CEO Rodrigo Nogueira, and it's stayed narrowly focused: Portuguese, Brazilian law, and Brazilian exams. Sabiá models are benchmarked against the OAB bar exam, ENEM, ENADE and USP/Unicamp entrance tests, and the newest Sabiá-4 Thinking reasoning model reportedly leads on some legal tasks at a fraction of frontier pricing.
The API is OpenAI-compatible, which makes swapping it into an existing stack straightforward, and pricing is transparent per-million-token in reais rather than hidden behind a sales call. The catch: everything is optimized for Brazil specifically, down to a 30% premium for guaranteed in-country inference. Outside Lusophone markets there's little reason to reach for it. As a sovereign-AI case study — small team, real government-data integration, exam-benchmarked accuracy — it's a legitimate model for the category, not just a marketing exercise.
What people say
Maritaca AI was founded in 2022 by researcher Rodrigo Nogueira together with AI researchers from Unicamp's studies center. The Sabiá model launched in October 2022, a month before ChatGPT's public debut, and the consumer-facing interface was renamed MariTalk in 2023.
The current lineup has three paid tiers plus a free trial: Sabiázinho-4 at R$1 per million input tokens / R$4 output, pitched as the fastest and cheapest option; Sabiá-3.1 at R$5 / R$10, positioned for general-purpose production use; and Sabiá-4 at R$5 / R$20, the most capable model for complex tasks. All three share a 128,000-token context window. Calling any model with a "-br-sp" suffix guarantees inference stays on Brazilian national territory, at a 30% price premium over the standard rate. Enterprise customers get custom billing and dedicated SLAs.
On performance, the company's own benchmarking has Sabiá-3 outperforming GPT models on Brazilian entrance and bar exams as of 2024, and Sabiá-4 Thinking is described as approaching frontier-model performance while leading on some legal-reasoning tasks at a fraction of the cost. A distinguishing feature is live access to Brazilian public data sources — the Central Bank, IBGE (statistics agency) and TSE (electoral court) — with the model citing its reasoning step by step rather than answering as a black box.
What's missing is independent third-party validation: no G2 or Capterra listing turned up, and no meaningful Reddit review thread either. Coverage is almost entirely Brazilian tech press (TechTudo, Alura, Data Hackers) recounting the company's own exam-score claims, plus a public maritalk-api GitHub repository with integration docs rather than user reviews. For buyers wanting benchmarking from outside the company itself, that's a real gap — the exam-beats-GPT claims are plausible given the localization work, but they're self-reported. Company-size trackers put the team at roughly 20 employees as of April 2026, a lean operation for a lab benchmarking itself against frontier-scale models from OpenAI and Google.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Maritaca AI (Sabiá)'s official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Maritaca AI (Sabiá) unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Maritaca AI (Sabiá) directly →
Spotted something out of date? Suggest an update →
More in LLMs & Chat
