Maritaca AI / Sabiá
Earlier Portuguese model offered through the retired API and a historically licensed MariTalk Local executable. Current private-license availability is unverified.
Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.
Report explicitly withholds architecture and training methodology.
Documented specialization, not a claim that other languages are unsupported.
No prompt-cache contract verified for this exact historical revision.
Historical executable requirements, not parameter counts. GPU memory may be distributed across multiple devices.
Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.
sabia-2-mediumEndpoint pricing has not been collected. Check the provider for applicable rates.
Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.
Not collectedNo production API selector verified.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.
Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.
The system processes eligible input and retains reusable state.
Reuse depends on matching input, session state and the cache policy.
Reusing state can reduce repeated work. Any charges follow the billing rules below.
sabia-2-mediumNo verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026Table 2: 38 exams held after mid-2023, not the full 64-exam suite. Provider evaluation; the medium model uses chat template disabled in this table.
Evaluation source (opens in a new tab)150 prompts in 13 categories; GPT-4 rates individual conversation turns on a ten-point scale, averaged across turns. Not the later pairwise win-rate metric.
Evaluation source (opens in a new tab)Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
Named provider model without a matching public weight release in the reviewed official inventory. Classification for historical artifacts is inferred; a public tokenizer, public upstream base or licensed executable is not a public release of this derivative.
Classification evidence (opens in a new tab)