Sabiá 4
Brazil's frontier Portuguese LLM — 128K context, January 2026 generation.
About
Released January 6, 2026 (checkpoint sabia-4-2026-01-06), Sabiá 4 is Maritaca AI's flagship large language model, purpose-built for Brazilian Portuguese. It offers a 128K-token context window with 32K max output tokens, and ships alongside BR-SP variants tuned for São Paulo-region deployment. Maritaca benchmarks it on Portuguese-native suites including PoETa v2 (44 tasks), the OAB Bar exam benchmark, and Sotaques Digitais — which tests irony, idioms, and regionalisms in everyday Brazilian writing. Older generations (Sabiá 2, early Sabiá 3 checkpoints) were deprecated in favor of the Sabiá 4 line.
Limitations: Portuguese-first — English and other languages are secondary capabilities. Available only through Maritaca's API and chat platform; no open weights. Pricing is in Brazilian reais, and BR-SP variants cost a ~30% premium. Benchmark leadership claims are on Portuguese-native evals, so cross-language comparisons should use those suites, not English leaderboards.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Vendor-reported on PoETa v2, OAB Bench, Sotaques Digitais (Portuguese-native suites)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 56; Trending score 65/100 from researcher assessment.