Tasks and inputs
Bulbul V4 Flash is listed for Text → Speech, Text-to-speech. It accepts text and produces audio. Use the publisher’s input format and preprocessing for your chosen deployment.
Bulbul V4 Flash is Sarvam's fast text-to-speech model for 11 Indian locales, with sub-250 ms first audio at a third less than Bulbul V3.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: INR 20 per 10,000 characters (Bulbul V3: INR 30). Indicative; confirm in your Sarvam account.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
If your voice product talks to customers in Tamil, Bengali or code-mixed Hinglish, the first question is cost per call. Bulbul V4 Flash answers it at INR 20 per 10,000 characters, a third below Bulbul V3's INR 30. Sarvam released it on 13 October 2026 and now recommends it for every new text-to-speech integration.
It is distilled from larger V4 research models into something small enough for production telephony: time to first audio is under 250 ms, and Sarvam says it runs up to 40% faster than V3 under comparable load.
Eleven locales are supported: Bengali, English, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil and Telugu (all -IN), plus code-mixed text. A real-time text normalisation layer decides how to speak things like PNR codes, PAN and Aadhaar numbers, dates and amounts, which is where many TTS systems stumble on Indian data.
The voice count depends on where you look. Sarvam's launch material lists 83 voices across 224 voice-and-style combinations; the API reference says more than 200 persona voices. These are probably two ways of counting the same catalogue. List voices from the API on your own key and use that inventory.
| Bulbul V3 | Bulbul V4 Flash | |
|---|---|---|
| Model ID | bulbul:v3 |
bulbul:v4-flash |
| Price | INR 30 / 10K chars | INR 20 / 10K chars |
| Request contract | Same as V3 | |
| Pitch, loudness | Not supported | |
| Pace | 0.5 to 2.0 | |
| Max per request | 2,500 characters |
Because the request contract is the same, switching is usually a model-ID change. Two details trip people up. First, remove pitch and loudness parameters. Second, preprocessing: the docs say it happens automatically for this model and also that enable_preprocessing is unsupported, so simply leave that flag out.
V4 research grew the training audio to more than 400 billion audio tokens, nearly seven times V3's speech data, with Indian-language representation as a primary objective. Those figures describe the V4 programme as a whole rather than Flash alone.
Customer care, sales and collections calls, banking and e-commerce notifications, education and narration are the intended uses. A more expressive Bulbul V4 Pro, able to laugh, sigh and sound nervous, has been previewed without a release date or price.
Write thirty sentences that mirror your real scripts, including account numbers, rupee amounts and mid-sentence English. Generate them with V3 and V4 Flash in each locale you serve and have native speakers rate them blind. Measure first-audio latency from your own region, not the vendor's. Compare against other text-to-speech models such as Gemini 3.8 Flash TTS if Indian languages are only part of your need, or see the Sarvam provider page.
Compare the model on the work you need to complete.
Bulbul V4 Flash is listed for Text → Speech, Text-to-speech. It accepts text and produces audio. Use the publisher’s input format and preprocessing for your chosen deployment.
Compare pronunciation, voice consistency and time to the first playable audio. Check the service’s billing unit for the actual script.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
inclusionAI
Ming-3.0-tts-0.5B is a compact open text-to-speech model from inclusionAI with voice design, zero-shot cloning and fine control over delivery.
ElevenLabs
Eleven v4 is ElevenLabs' flagship text-to-speech model for audiobooks, dubbing and emotional dialogue, with 90+ languages and 10,000 characters per request.
ElevenLabs
Eleven v4 Turbo is ElevenLabs' low-latency v4 speech model for voice agents, with about 100 ms median inference, 90+ languages and half v4's price.