Doubao
ByteDance's multimodal AI assistant for chat, images, and voice - China's most-used AI app.
India-built sovereign LLMs and voice APIs tuned for Indian languages
Sarvam is one of the most credible names in Indian-language AI, with founders who have real infrastructure pedigree and a rare government mandate to build a sovereign LLM. The genuinely interesting angle is Indic coverage and voice, not frontier-model bragging rights, so evaluate it against your specific languages rather than global leaderboards. Pricing is usage-based per token and has reportedly changed more than once, so check the live API pricing page before you budget. Model names and sizes (Sarvam-M, 30B, 105B) are moving targets, and some capabilities are newer and less battle-tested than incumbents. For teams that need Hindi, Tamil, Telugu and other Indian languages done well, it is worth a serious pilot.
Sarvam AI is a Bengaluru-based startup building sovereign, Indian-language large language models and speech APIs. Selected under the IndiaAI Mission to build a national foundational LLM, it offers models (Sarvam-M, Sarvam-30B, Sarvam-105B) plus ASR, TTS and translation APIs aimed at developers and enterprises serving India's many languages.
Sarvam AI was founded in August 2023 in Bengaluru by Dr. Vivek Raghavan (a former architect of the Aadhaar programme) and Dr. Pratyush Kumar (previously at Microsoft Research and AI4Bharat). Its stated mission is a full Indian-language AI stack: foundational text models, voice-first interfaces and developer APIs designed for the country's many languages and lower-cost hardware. In April 2025 it was reportedly selected first from 67 applicants under the IndiaAI Mission to build India's sovereign LLM, with government support that reportedly included a subsidy toward acquiring thousands of NVIDIA H100 GPUs. The product line spans several models. Sarvam-M is a reported 24B-parameter instruction-tuned model (a derivative of Mistral-Small) released in May 2025 and post-trained on English plus several Indic languages; Sarvam-30B is an open-weights model reported to have launched in early 2026; and Sarvam-105B is a larger flagship. On top of the models, Sarvam offers speech-to-text (ASR), text-to-speech (TTS), translation and chat APIs through its developer platform, positioned for enterprises and builders who need Indian-language support. Sarvam has raised institutional funding, including a reported $41M Series A in late 2023 (Lightspeed, Peak XV, Khosla Ventures) and a Series B first close reported at $234M that put it at a reported $1.5B valuation, with HCLTech named as a strategic lead investor. As with any fast-moving startup, treat specific model sizes, prices and funding figures as vendor or press claims that can change; verify the current numbers on Sarvam's own docs before relying on them.
Sarvam AI is a Bengaluru startup building sovereign, Indian-language LLMs and speech APIs, selected under the IndiaAI Mission to build a national foundational model. Its edge is Indic-language and voice coverage for developers and enterprises, not global-benchmark supremacy.
Founded in August 2023 in Bengaluru by Dr. Vivek Raghavan and Dr. Pratyush Kumar, Sarvam AI develops foundational LLMs (Sarvam-M, Sarvam-30B, Sarvam-105B) and speech/translation APIs for Indian languages. In 2025 it was reportedly chosen first among 67 applicants under the IndiaAI Mission to build India's sovereign LLM, with government-backed GPU support.
Chat/LLM APIs, automatic speech recognition (ASR), text-to-speech (TTS) with Indian voices, and machine translation across Indian languages. Some models are released as open weights for self-hosting. APIs are reported to include OpenAI-compatible endpoints, and models are distributed via Hugging Face and aggregators such as OpenRouter.
Developers, startups and enterprises in India (and those serving Indian-language users abroad) who need high-quality Indic language, speech and translation support, plus public-sector and regulated buyers who value an India-hosted, sovereign AI stack.
Developers and ML engineers integrating Indian-language chat, voice or translation into apps.
Startup CTOs, enterprise AI/platform leads and public-sector technology teams choosing an Indic AI provider.
AI4Bharat and Indic-NLP researchers, IndiaAI Mission stakeholders, and enterprise architects evaluating sovereign AI options.
An India-focused product or enterprise team that must serve users in multiple Indian languages with strong speech and translation, and that values a locally built, sovereign stack over a global general-purpose model.
Reported $41M Series A in late 2023 led by Lightspeed with Peak XV Partners and Khosla Ventures. A Series B first close was reported at $234M at a reported $1.5B valuation, with HCLTech named as a strategic lead investor and Bessemer Venture Partners among participants. Treat all figures as reported and verify current status.
Sarvam AI is a Bengaluru-based startup building large language models and speech APIs optimised for Indian languages. It was reportedly selected under the government's IndiaAI Mission to build a sovereign foundational LLM for India.
Sarvam focuses on Indian languages, with models and speech APIs reported to cover English plus a range of major Indic languages such as Hindi, Tamil, Telugu, Bengali, Marathi and others. Check the docs for the current supported list.
Sarvam is a developer/API product with usage-based pricing and reportedly offers free credits to start. Some models are released as open weights that you can self-host at your own cost. Verify current pricing on the official site.
It was founded in August 2023 by Dr. Vivek Raghavan and Dr. Pratyush Kumar, who bring backgrounds in India's digital-identity infrastructure and Indic AI research respectively.
Side-by-side pages for pricing, features, and best-fit use cases.
ByteDance's multimodal AI assistant for chat, images, and voice - China's most-used AI app.
Alibaba's free AI assistant, backed by the open-weight Qwen model family
Free AI chat assistant from Moonshot AI, built for long-context reasoning and open Kimi K2 models
Meta's open-weight 30B agentic model that runs local, multimodal AI agents on a single GPU