Sarvam-M
24B open-weights hybrid model for Indian languages and reasoning
About
Sarvam-M is a 24-billion-parameter open-weights hybrid model from Bengaluru's Sarvam AI, built on Mistral Small and released on May 23, 2025. It was tuned with supervised fine-tuning, reinforcement learning with verifiable rewards, and FP8 inference optimisation, with roughly a third of training prompts in ten Indian languages. The company claims strong gains on Indic benchmarks and math over the base model, though English knowledge benchmarks barely move. Independent coverage has questioned real-world adoption, so treat the benchmark claims as company-reported.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Company-reported: +20% on Indic language benchmarks, +21.6% on math and +17.6% on programming over the Mistral Small base; 86% gain on romanised GSM-8K; on par with Llama-3.3 70B but about a point behind on English MMLU.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Selected by IndiaAI Mission for sovereign LLM work; strong press coverage; Benchmark figures are company-reported; independent outlets questioned adoption; Built on Mistral Small with documented SFT/RLVR pipeline. License: License not confirmed in sources read this session; derived from Mistral Small (Apache 2.0). Confirm on the model card before commercial use..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Selected by IndiaAI Mission for sovereign LLM work; strong press coverage
- · Benchmark figures are company-reported; independent outlets questioned adoption
- · Built on Mistral Small with documented SFT/RLVR pipeline