Sarvam 30B
Sarvam 30B is a 30B parameter Mixture-of-Experts chat and reasoning model from Sarvam AI, optimized for Indian languages, real-time conversation, high-throughput voice-agent pipelines, coding, and practical deployment.
Provider
Sarvam AI
Model family
Sarvam
Chat LLM
Cost tier
30b
Status
Current
Release Mar 6, 2026
Why teams choose it
Complex reasoning
Useful for workflows that require structured thinking, multi-step logic, and deeper analysis than lightweight models provide.
Long-context analysis
Helps teams summarize, compare, and extract insights from long documents without losing important nuance.
Sarvam AI roadmap vigilance
Use published model pages—not stale marketing blurbs—for modalities, quotas, pricing, and policy; schedule revalidation tied to vendor release notes.
Cost-efficient routing
Useful as part of a routing stack where cheap models handle drafts and confirmations and this tier handles genuinely hard passages.
Tradeoffs to know
- Use Sarvam 105B when maximum reasoning quality matters more than deployment efficiency.
- Published benchmark results are vendor-reported and need local evaluation.
- Thinking mode can consume output budget unless max_tokens and reasoning settings are tuned.
When not to use this
- Self-hosting outcomes depend on hardware, quantization, and ops maturity—budget time beyond swapping an API hostname.
- May demand more instrumentation than SaaS-managed APIs to duplicate latency, failover, and support guarantees.
- Benchmark prompts and regressions continuously before rewriting entire routing tables around weights.
Technical specs
- Inputs
- text
- Outputs
- text
- Capabilities
- Indian-language chat, Real-time conversation, Reasoning, Coding, Voice-agent pipelines, Tool calling, OpenAI-compatible chat completions
- License
- Apache 2.0
- Model string
sarvam-30b
Benchmarks
{
"mbpp": 92.7,
"mmlu": 85.1,
"source": "https://www.sarvam.ai/blogs/sarvam-30b-105b",
"math500": 97,
"mmlu_pro": 80,
"tau2_avg": 45.7,
"aime_2025": 88.3,
"humaneval": 92.1,
"browsecomp": 35.5,
"vendor_reported": true,
"live_code_bench_v6": 70,
"aime_2025_with_tools": 96.7,
"indian_language_win_rate_avg": "89%"
}When to choose Sarvam 30B instead of Sarvam 105B
Sarvam positions the 30B model as the lower-latency, lower-cost option for real-time conversational workloads, while Sarvam 105B prioritizes quality for complex reasoning, coding, and long-form generation.
- Sarvam 30B has a 64K context window; Sarvam 105B has a 128K context window.
- Route simple chat and voice-agent turns to 30B only after measuring language quality and escalation accuracy.
- Keep a 105B fallback for tasks that exceed the 30B quality threshold in your evaluation set.
Sources: Sarvam chat model selection
Sarvam family lineup
Current models
Compare with
Sarvam 30B FAQ
What is Sarvam 30B?
Sarvam 30B is a 30B parameter Mixture-of-Experts chat and reasoning model from Sarvam AI, optimized for Indian languages, real-time conversation, high-throughput voice-agent pipelines, coding, and practical deployment. Sarvam documents 2.4B active parameters per token, 16T tokens of pre-training data, a 64K context wi...
When does Sarvam 30B fit best?
High-throughput Indian-language chat and support assistants.
What should teams watch out for with Sarvam 30B?
Use Sarvam 105B when maximum reasoning quality matters more than deployment efficiency.
Explore next
Models, tools, and comparisons that connect to this reference.