Exact numeric context from the documented model-list response, corroborated by the model catalog. Shared context, not an additional independent input allowance.
Provider-wide Responses API JSON-schema support, with examples using sabia-4. Subset only: up to 100 object properties, five nesting levels, 15,000 total schema-name characters and 500 enum values. Not all JSON Schema constraints are supported.
Rates published for this model and BR-SP. One model per JSONL file, at most 50,000 requests / 200 MB, 24-hour completion window; completed output files are kept for 30 days. Expiration is possible.
Exact base checkpoint and dense/MoE configuration are not identified. The training blog’s trillion-parameter cost exercise is illustrative, not this model’s specification.
Prices retain their published currencies and billing units below.
Published billing units · original currency · observed 14 Oct 2026
Category
Usage
Rate
Unit
Standard
Input1
R$5.00
1M tokens
Standard
Output1
R$40.00
1M tokens
Caching
Cached input1
R$1.25
1M tokens
Night
Input1
R$3.50
1M tokens
Night
Output1
R$28.00
1M tokens
Flex
Input1
R$2.50
1M tokens
Flex
Output1
R$20.00
1M tokens
Batch
Input1
R$2.50
1M tokens
Batch
Output1
R$20.00
1M tokens
Tool add-on
Web search2
R$0.0165
search
Tool add-on
Page reading2
R$0.066
page read
Tool add-on
Code execution2
R$0.016
minute, rounded up
Tool add-on
Data Ocean2
R$0.10
GB processed
File extraction add-on
PDF extraction / medium3
R$0.02
PDF page
File extraction add-on
PDF extraction / advanced3
R$0.045
PDF page
Rate conditions
BRL per million tokens. Cache hits retain their own 75% input discount. Batch, Flex and night rates apply only to uncached input and output and are mutually exclusive. Night hours are 22:00–06:00 BRT. Flex is capacity-dependent and may queue up to five minutes or return 429. No currency conversion is applied.
Additional server-side tool fee, separate from token charges. Integrated tools require stream:false. A web-search request may perform both searches and page reads. Data Ocean can also trigger search, reading and code fees. No BR-SP tool-fee premium is specified.
Additional charge per PDF page, separate from model tokens. Repeated extraction can be billed again; keep the returned extracted text for subsequent requests. The level setting does not apply to other file types.
Provider says API inputs/outputs are not used for model training. Normal content is discarded after generation, but error/observability logs may retain request content up to 30 days and infrastructure logs up to 18 months. Batch output files have a separate 30-day retention period; prompt-cache lifetime is unspecified. Chat/app conversations have different storage rules.
Prices retain their published currencies and billing units below.
Published billing units · original currency · observed 14 Oct 2026
Category
Usage
Rate
Unit
Standard
Input1
R$6.50
1M tokens
Standard
Output1
R$52.00
1M tokens
Caching
Cached input1
R$1.63
1M tokens
Night
Input1
R$4.55
1M tokens
Night
Output1
R$36.40
1M tokens
Flex
Input1
R$3.25
1M tokens
Flex
Output1
R$26.00
1M tokens
Batch
Input1
R$3.25
1M tokens
Batch
Output1
R$26.00
1M tokens
Tool add-on
Web search2
R$0.0165
search
Tool add-on
Page reading2
R$0.066
page read
Tool add-on
Code execution2
R$0.016
minute, rounded up
Tool add-on
Data Ocean2
R$0.10
GB processed
File extraction add-on
PDF extraction / medium3
R$0.02
PDF page
File extraction add-on
PDF extraction / advanced3
R$0.045
PDF page
Rate conditions
BRL per million tokens. Cache hits retain their own 75% input discount. Batch, Flex and night rates apply only to uncached input and output and are mutually exclusive. Night hours are 22:00–06:00 BRT. Flex is capacity-dependent and may queue up to five minutes or return 429. No currency conversion is applied. BR-SP token charges include the published 30% premium; cached-input figures are the rounded published values, not an inferred higher-precision rate.
Additional server-side tool fee, separate from token charges. Integrated tools require stream:false. A web-search request may perform both searches and page reads. Data Ocean can also trigger search, reading and code fees. No BR-SP tool-fee premium is specified.
Additional charge per PDF page, separate from model tokens. Repeated extraction can be billed again; keep the returned extracted text for subsequent requests. The level setting does not apply to other file types.
Provider says API inputs/outputs are not used for model training. Normal content is discarded after generation, but error/observability logs may retain request content up to 30 days and infrastructure logs up to 18 months. Batch output files have a separate 30-day retention period; prompt-cache lifetime is unspecified. Chat/app conversations have different storage rules.
BRL per million tokens. Cache hits retain their own 75% input discount. Batch, Flex and night rates apply only to uncached input and output and are mutually exclusive. Night hours are 22:00–06:00 BRT. Flex is capacity-dependent and may queue up to five minutes or return 429. No currency conversion is applied.
Observed 14 Oct 2026. Write rates and storage fees are different charges; use the billing rules above.
BRL per million tokens. Cache hits retain their own 75% input discount. Batch, Flex and night rates apply only to uncached input and output and are mutually exclusive. Night hours are 22:00–06:00 BRT. Flex is capacity-dependent and may queue up to five minutes or return 429. No currency conversion is applied. BR-SP token charges include the published 30% premium; cached-input figures are the rounded published values, not an inferred higher-precision rate.
Observed 14 Oct 2026. Write rates and storage fees are different charges; use the billing rules above.
Caching reviewed 14 Oct 2026
What the measurements say
Agent category mean
Maritaca AI · vendor-reported
94provider score / 100
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
June 2026 launch suite. Simple category means, then three-category overall mean, not an unweighted mean of all 11 tasks. Mixed metrics: Pix/Ticket pass@1; MARCA uses GPT-4.1 checklist judging; OAB compares human scores; legal drafting and Sotaques use GPT-5.4 judging; BLUEX uses Sabiazinho 4 answer judging; ENAMED/POSCOMP exact match; PoETa v2 NPM over 44 tasks and ten runs per task. Sotaques originally uses a 1–5 judge scale, reported here as normalized percentage. Includes private benchmarks and is not an independent ranking.
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.
The record behind the model
Why it is included
Provider explicitly states Sabiá 4 models are API-only and offers neither weight downloads nor local/on-premise licensing.
