What’s on your mind?
Think it through. Find the words. Make it happen.
A thought. Many languages.
Translate meaning, change scripts, or find the language.
Good ideas deserve a voice.
Turn words into audio, or capture what’s being said.
There’s more between the lines.
Read a scan, extract the details, or translate a whole document.
Your voice. A wider world.
Create a voice from a reference clip, or localize your audio and video.
Uses standard voices. Voice cloning is off for dubbing.
Model Reference & Official Pricing
Comprehensive A-to-Z breakdown of Sarvam AI model architectures, context windows, token limits, and consumption pricing in Indian Rupees (₹ INR). Every new account receives ₹100 in free credits upon registration at dashboard.sarvam.ai.
| Service / Model | Model Identifier | Context / Billable Unit | Input Rate (INR) | Cached Input | Output / Generation | Status |
|---|---|---|---|---|---|---|
| Sarvam 105B | sarvam-105b |
128,000 tokens (128K) | ₹15 / 1M tokens | ₹5 / 1M tokens | ₹60 / 1M tokens | GA · Live |
| Sarvam 105B · Fast | sarvam-105b-conversations |
32,000 tokens (32K) | ₹15 / 1M tokens | ₹5 / 1M tokens | ₹60 / 1M tokens | GA · Live |
| DeepSeek v4.1 Flash | deepseekv4.1-flash |
1,048,576 tokens (1M+) | ₹25 / 1M tokens | ₹0.50 / 1M tokens | ₹105 / 1M tokens | Beta (Whitelist) |
| GLM 5.3 | glm5.3 |
1,048,576 tokens (1M+) | ₹126 / 1M tokens | ₹20 / 1M tokens | ₹390 / 1M tokens | Beta (Whitelist) |
| Gemma 4 | gemma4 |
128,000 tokens (128K) | ₹35 / 1M tokens | ₹10 / 1M tokens | ₹90 / 1M tokens | Beta (Whitelist) |
| Bulbul v3 TTS | bulbul:v3 |
10,000 characters | — | — | ₹30 / 10K chars | GA · Live |
| Voice Cloning | voices/clone |
10,000 characters | 10–15s audio clip | — | ₹30 / 10K chars | GA · Live |
| Saaras v3 STT | saaras:v3 |
Per Audio Hour | ₹30 / hour | — | ₹45 / hr (with Diarization) | GA · Live |
| Speech Translation | speech-to-text-translate |
Per Audio Hour | ₹30 / hour | — | ₹45 / hr (with Diarization) | GA · Live |
| Sarvam Translate v1 | sarvam-translate:v1 |
10,000 characters | — | — | ₹20 / 10K chars | GA · Live |
| Transliteration | transliterate |
10,000 characters | — | — | ₹20 / 10K chars | GA · Live |
| Language Identification | text-lid |
10,000 characters | — | — | ₹3.50 / 10K chars | GA · Live |
| Document Digitization | document-digitization |
Per Page (Vision OCR) | — | — | ₹0.50 / page | GA · Live |
| Document Extraction | document-extraction |
Per Page (Schema JSON) | — | — | ₹1.00 / page | GA · Live |
Sarvam 105B
sarvam-105bFlagship Indic large language model trained on deep Indian cultural contexts, formal records, and conversational idioms. Reasoning is enabled by default to tackle multi-step problems accurately.
Sarvam 105B · Fast
sarvam-105b-conversationsSpecialized variant fine-tuned for rapid turn-taking, concise responses, and conversational voice agents where low latency and direct answers are essential.
DeepSeek v4.1 Flash
deepseekv4.1-flashHigh-performance open-weight model with a massive 1M+ token context window. Delivers live chain-of-thought reasoning traces, math problem solving, and long document synthesis.
GLM 5.3
glm5.3Advanced software engineering and autonomous agent model. Engineered for complex code generation, long-horizon multi-step tool use, and rigorous architectural synthesis.
Gemma 4
gemma4Lightweight, efficient open-weight multilingual reasoning model offering rapid throughput and economical token rates for standard question-answering and classification.
Bulbul v3 (TTS)
bulbul:v3Expressive Indian text-to-speech engine supporting 11+ Indic languages with 6 distinct speaker voices and natural conversational prosody.
Voice Cloning
voices/cloneZero-shot voice cloning from a 10–15 second audio reference clip. Synthesize custom speech across languages in your own reference voice.
Saaras v3 (STT)
saaras:v3Robust Indic automatic speech recognition trained to transcribe accented, colloquial, and code-mixed speech with exceptional fidelity.
Sarvam Translate v1
sarvam-translate:v1High-accuracy neural machine translation preserving semantic intent, idioms, and colloquial expressions across 22 official Indian languages and English.
Document Intelligence
document-digitizationLayout-aware document vision OCR and schema extraction for scanned forms, invoices, receipts, and handwritten documents in Indic languages.
Preset Support Email to Activate Open-Source Models
Sarvam AI's V2 API models (deepseekv4.1-flash, glm5.3, gemma4) require account whitelisting. Use this pre-formatted email to request instant beta access from Sarvam Support.
