Sarvam AI

Indian-language speech recognition, speech synthesis, translation and document intelligence, with managed and customer-hosted deployment options.

14 reviewed release and service records, including 222 Bulbul v4 Flash personas. Public-weight language models are excluded. Historical versions, ambiguous conversational-model licensing and some deployment details remain under review.

Closed-weight models

Browse by model typeCapabilities can overlap

Start with the job your model needs to do. Then narrow by capabilities, provider and price.

Select up to 3 models to compare

What does one question cost?

Compare every model with the same request.

Input tokensOutput tokens

Estimate = (input tokens × input price + output tokens × output price) / 1,000,000. Include conversation history in input and billed reasoning in output. Input-only decision models ignore the output setting; enter all billed input for one decision, including its question and criteria. Media, tools and discounts are separate.

14 models

Sort by

Model input and output limits, prices and estimated cost per question at the selected token counts.
Input tokensPublished limit*	Output tokensMaximum per response*	Input pricePer 1M tokens	Cost per question1,000 input + 500 outputEstimated · uncached
Not collected	Not collected	Usage pricingNot collected	Different billing unitPrice details ↗
1,500 characters per REST requesttext input	Not collected	Usage pricingNot collected	Different billing unitPrice details ↗
2,500 characters per REST requesttext input	WAV, MP3, AAC, FLAC, Linear16output formats	Usage pricing ₹30.00 / 10K charactersSpeech synthesis · Sarvam APIRates & conditions ↗	Different billing unitPrice details ↗
2,500 characters per REST requesttext input	WAV, MP3, AAC, FLAC, Linear16output formats	Usage pricingNot collected	Different billing unitPrice details ↗
REST up to 30 seconds; legacy batch duration not verifiedaudio limit	Not collected	Usage pricing ₹30.00 / audio hourSpeech to English translation · Sarvam API₹45.00 / audio hourSpeech recognition with diarization · Sarvam APIRates & conditions ↗	Different billing unitPrice details ↗
REST up to 30 seconds; batch up to 2 hours per fileaudio limit	Not collected	Usage pricing ₹30.00 / audio hourSpeech recognition / translation · Sarvam API₹45.00 / audio hourSpeech recognition with diarization · Sarvam APIRates & conditions ↗	Different billing unitPrice details ↗
Not collected	Not collected	Usage pricing ₹30.00 / audio hourSpeech recognition / translation · Sarvam APIRates & conditions ↗	Different billing unitPrice details ↗
REST up to 30 seconds; batch up to 2 hours per fileaudio limit	Not collected	Usage pricing ₹30.00 / audio hourSpeech recognition / translation · Sarvam API₹45.00 / audio hourSpeech recognition with diarization · Sarvam APIRates & conditions ↗	Different billing unitPrice details ↗
30 seconds REST; up to 2 hours per batch fileaudio limit	Not collected	Usage pricing ₹30.00 / audio hourSpeech recognition · Sarvam API₹45.00 / audio hourSpeech recognition with diarization · Sarvam APIRates & conditions ↗	Different billing unitPrice details ↗
1,000 characters per requesttext input	Not collected	Usage pricing ₹20.00 / 10K charactersText translation · Sarvam APIRates & conditions ↗	Different billing unitPrice details ↗
1,000 REST/HTTP stream; 2,500 per WebSocket text messagetext input	WAV, MP3, FLAC, Opus, AAC, Linear16, μ-law, A-lawoutput formats	Usage pricing ₹30.00 / 10K charactersCloned speech synthesis · Sarvam APIRates & conditions ↗	Different billing unitPrice details ↗
Not collected	Not collected	Usage pricingNot collected	Different billing unitPrice details ↗
Not collected	Not collected	Usage pricingNot collected	Different billing unitPrice details ↗
Not collected	Not collected	Usage pricing ₹0.50 / pageDocument digitization · Sarvam API₹1.00 / pageDocument field extraction · Sarvam APIRates & conditions ↗	Different billing unitPrice details ↗

* Media models show their native limits. A shared context window includes both input and output tokens.

Original currency and billing unit, without currency conversion. Rates are dated snapshots; open the conditions for tiers, promotions and endpoint restrictions.How to read the data

Catalog coverage

Closed-weight model variants and explicitly identified closed-source managed model components across language, code, image, video, audio and other documented modalities. Global provider and historical-release coverage is being expanded; this is not yet exhaustive. 172 company reviews are still open.

See what is being reviewed