The pitch from Sarvam AI is straightforward, perhaps deceptively so. Apply to their new startup program, and if selected, you'll receive six to twelve months of free access to the Bengaluru company's language models—along with engineering support and some marketing help. It's an opening gambit that dozens of AI companies have tried before, from Google to Anthropic to Microsoft. But Sarvam's angle is different, and it matters: native support for 22 Indian languages, including the kind of code-switching and accent recognition that engineers building for India's vast multilingual market have been quietly begging for.
The company launched the program on March 5, timing that suggests urgency. Just weeks earlier, on February 18, Sarvam had unveiled a flurry of new models at the India AI Impact Summit—30-billion and 105-billion parameter releases, trained with compute from the IndiaAI Mission and infrastructure from Yotta. They'd also rolled out Samvaad, a conversational platform aimed at enterprises, and Arya, what they're calling an "agent orchestration stack." In other words, Sarvam isn't just building language APIs. They're assembling a full operational stack for companies trying to crack the Indian market.
Whether this program gains traction will depend on something harder to predict: whether India's AI startup founders see enough value in language-specific tooling to bet on a younger player over the established cloud giants. The credits don't roll over. The exact amounts aren't publicly disclosed—Sarvam says allocation depends on your use case and stage, which is either flexible or opaque depending on your perspective. And when the free period ends, you're on standard pricing.
Still, for startups wrestling with Tamil customer support or trying to parse Hinglish conversations or extract handwritten Gujarati text from scanned invoices, the major cloud providers often stumble. That's the gap Sarvam is betting on.
The Offer, Unpacked
The program targets a specific cohort: startups building products around speech, translation, or chat. You apply through Sarvam's website, and if approved, you gain access to their production APIs without immediate billing concerns. According to the official program page, credits are "matched to your scale" over the program's duration. How much that actually means varies—applicants are asked to estimate their projected API consumption for years one and two, which suggests Sarvam is calibrating allocations case by case.
The API suite itself is comprehensive, if not always clearly priced. Speech-to-text runs through models called Saaras and Saarika, covering 22 Indian languages—Hindi, Bengali, Tamil, Telugu, Gujarati, Kannada, Malayalam, Marathi, Punjabi, Odia, and more—plus English with Indian accent recognition. The system handles code-mixing, which matters considerably when users toggle between languages mid-sentence, as many do. Automatic language detection is included.
Text-to-speech offers more than 30 voices across 11 languages, with controls for pacing and emotional temperature. There's Bulbul v3 and v2, depending on what you need. Document intelligence, powered by what Sarvam calls Vision 3B, extracts text and tables from scanned documents in 23 languages. All of it's accessible via REST, with streaming options for the speech models.
Beyond the credits, participants get direct access to Sarvam's engineering team and the promise of co-branded case studies for launch visibility. Business Today noted that the program aims to support "AI builders in India," though the application form doesn't explicitly restrict geographic eligibility—a subtle indication that Sarvam may be casting a wider net than the marketing suggests.
What It Actually Costs
Standard pricing for Sarvam's APIs runs ₹30 per hour for speech-to-text, ₹20 per 10,000 characters for translation, and ₹15 to ₹30 per 10,000 characters for text-to-speech, depending on which model version you're using. The chat model, Sarvam-M, is currently listed at ₹0 per token—a promotional rate that presumably won't last forever, though the company hasn't said when it might change.
Document intelligence pricing is murkier. Some parts of the documentation mark it as "currently free," while earlier references mention "Free for February," which suggests either outdated copy or a shifting pricing strategy. Either way, confirming current terms directly with Sarvam before committing is probably wise.
Speech-to-text operates with median latency under 250 milliseconds, according to the product page, and supports both batch and streaming modes. Diarization—speaker separation—is available as an add-on. Rate limits vary by tier: 60 requests per minute on the Starter plan, 200 on Pro, 1,000 on Business. It's a tiered structure that mirrors most API providers, though the exact rate limits may shift as the company scales.
The Fine Print and the Competition
Program participants must display "Powered by SarvamAI" branding in their products. A reasonable ask, though one that might give some founders pause if they're building consumer-facing apps where third-party branding disrupts the experience. Credits expire when the program term ends—no banking them for later. And when the free period concludes, you're on the hook for standard pricing, which could be a jarring transition for startups that scaled aggressively during the free window.
Existing Sarvam API users can apply, which suggests the company is trying to convert trial users into deeper partnerships as much as attract entirely new customers.
The application itself is thorough. Company information, product stage (development, beta, production), funding stage (pre-seed through Series B+, or bootstrapped), industry, team size, founder details with LinkedIn profiles, which APIs you're interested in, projected usage scales, a demo link, and referral information if someone sent you their way. Sarvam hasn't disclosed cohort size, acceptance rates, or how long decisions take. That opacity could be strategic—or it could simply mean they're still figuring out the mechanics.
For context, Google Cloud's AI Startup Program offers up to $250,000 in credits for the first year, with an additional $100,000 in year two at 20 percent usage. Google for Startups Accelerator: AI First, which specifically targets Indian startups, provides up to $350,000 in Google Cloud credits over a three-month equity-free program, though eligibility requirements apply. Anthropic for Startups distributes API credits and rate limit increases through partner VCs, with terms varying by partner—a structure that favors venture-backed companies over bootstrapped ones. Microsoft's Founders Hub discontinued OpenAI credits partway through last year, according to support forum discussions, a reminder that these programs aren't always permanent.
Sarvam's approach isn't about competing on credit size. They're betting on language capability as the differentiator. If your product genuinely needs to work in Tamil, or handle Hinglish code-switching, or process handwritten Gujarati invoices, the major cloud providers' models frequently fall short. Sarvam's 22-language support is purpose-built for that gap, trained on data that reflects how people in India actually communicate—not just formal written Hindi or English, but the messy, multilingual reality of everyday speech.
Whether that's enough depends on what you're building. And whether Sarvam can sustain this level of support as it scales. The program launched just weeks after the company's flurry of model releases, which could signal confidence—or it could mean they're racing to build adoption before competitors close the language gap themselves.
Applications are open now through the Startup Program page at sarvam.ai. For startups already committed to multilingual Indian markets, it's worth a look. For those still weighing whether language diversity is central to their product strategy, the decision is less obvious. But that's precisely the market Sarvam is trying to define.
