Pular para o conteúdo principal

Models

Maritaca AI offers a range of models in the Sabiá family, designed to meet various needs and usage scenarios. Each model is adjusted to provide an optimal balance between intelligence, speed, and cost, allowing you to choose the option that best aligns with the requirements of your project.

Available models​

S5​

Our most intelligent model, recommended for analyzing documents, working with data, and solving complex problems. Use it to build assistants that look up information and perform tasks with tools.

Example use cases: Document analysis and summarization, information extraction, task automation with tools, and structured content generation.

API details: The reasoning_effort parameter controls reasoning intensity (none, low, medium or xhigh, the default). S5 supports up to 1 million tokens of context and 64K output tokens, and is also available through the Batch API. API model names: S5 or s5.

S5 Free​

S5 free of charge for a limited time, to try the model at no cost. Text only; Batch API, Flex and built-in tools are not included, and rate limits are fixed (see Rate Limits). API model names: S5-free or s5-free.

Available for a limited time

S5 Free is a temporary launch offer and may be discontinued at any time. For ongoing or production use, use S5.

S5 BR-SP​

The same S5 model, with inference and processing performed entirely (100%) within Brazilian territory — intended for use cases that require Brazilian data sovereignty and residency. Capabilities, context, and training data are identical to S5; pricing is 30% higher. API model names: S5-br-sp or s5-br-sp.

S5 Nano​

A compact version of S5, with open weights (maritaca-ai/S5-nano on Hugging Face) and the lowest cost in the family: R$ 0.10 per 1M input tokens and R$ 0.60 per 1M output tokens. It has the same native reasoning as S5 and, as an open model, returns its reasoning as plain text, unencrypted (see Reasoning). It supports up to 256K tokens of context, function calling, built-in tools, and the Batch API. Accepted API names: S5-nano or s5-nano.

Use cases: High-volume classification and information extraction, customer support and chat with fast responses, routing and triage in agentic flows, and simple tasks where cost and latency matter more than peak quality.

Previous-generation models​

Sabiá 4 Thinking​

The reasoning model of the Sabiá family, delivering frontier quality in Portuguese and Brazilian contexts at the lowest cost among the evaluated models. Compared to Sabiá 4, it brings significant gains in tool use, legal tasks, and answer quality. Recommended for complex tasks that benefit from explicit reasoning before answering. Supports up to 1 million tokens of context, so long documents, entire case files or whole knowledge bases fit in a single request (also available through the Batch API).

Example use cases: Multi-step problem solving, agentic workflows and tool use, legal analysis and drafting, and exam questions and technical reasoning.

Sabiá 4 Thinking BR-SP​

The same Sabiá 4 Thinking model, with inference and processing performed entirely (100%) within Brazilian territory — intended for use cases that require Brazilian data residency. Capabilities, context, and training data are identical to Sabiá 4 Thinking; pricing is 30% higher. API model name: sabia-4-thinking-br-sp.

Sabiá 4​

General-purpose model focused on agentic capability and Brazilian context, recommended for uses that need accuracy with cost efficiency. Read the technical paper (arXiv) for more details.

Example use cases: Drafting and reviewing briefs and opinions, tutoring for exams, analyzing and synthesizing long documents, and orchestrating multi-step agentic workflows.

Sabiá 4 BR-SP​

The same Sabiá 4 model, with inference and processing performed entirely (100%) within Brazilian territory — intended for use cases that require Brazilian data sovereignty and residency. Capabilities, context, and training data are identical to Sabiá 4; pricing is 30% higher. API model name: sabia-4-br-sp.

Sabiazinho 4​

Optimized for large-scale applications with a focus on speed, low cost, and solid accuracy. Delivers the best performance in agentic workflows and is recommended for information extraction, educational assessments, and legal applications, including awareness of the Brazilian legal context.

Example use cases: Drafting and reviewing legal documents, analyzing contracts, applying and interpreting Brazilian law, classifying and extracting information from documents, supporting large-scale educational assessments, and running agentic multi-step automations with fast responses and lower cost.

Sabiazinho 4 BR-SP​

The same Sabiazinho 4 model, with inference and processing performed entirely (100%) within Brazilian territory — intended for use cases that require Brazilian data sovereignty and residency. Capabilities, context, and training data are identical to Sabiazinho 4; pricing is 30% higher. API model name: sabiazinho-4-br-sp.

Specifications​

S5S5 FreeS5 BR-SPS5 Nano
Max context1M1M1M256K
Max output tokens64K64K64K32K
Model names/aliasesS5
s5
S5-free
s5-free
S5-br-sp
s5-br-sp
S5-nano
s5-nano

Previous-generation models​

Sabiá 4 ThinkingSabiá 4 Thinking BR-SPSabiá 4Sabiá 4 BR-SPSabiazinho 4Sabiazinho 4 BR-SP
Max context1M1M128K128K128K128K
Max output tokens32K32K32K32K32K32K
Model names/aliasessabia-4-thinkingsabia-4-thinking-br-spsabia-4
sabia-4-2026-01-06
sabia-4-br-spsabiazinho-4
sabiazinho-4-2026-01-06
sabia-4-small
sabiazim-4
sabiazinho-4-br-sp

For pricing information, see the Pricing page.

Deprecated models​

As we release more advanced and secure models, we regularly discontinue older models. In the table below, you will find a complete list of deprecated models, along with recommended replacements.

ModelDeprecation dateRecommended replacement
sabia-2-medium2024-10-01sabiazinho-4
sabia-2-small2025-01-15sabiazinho-4
sabia-3-2024-07-152024-10-01sabiazinho-4
sabia-3-2024-09-092025-02-17sabiazinho-4
sabiazinho-3-2024-11-132025-03-06sabiazinho-4