Model Details
Sarvam 30B
Quick answer: Sarvam 30B is Sarvam AI's July 2025 Indian-origin 30B model scoring 92.1% HumanEval, 80% MMLU-Pro, and 49.0% Arena Hard v2. Apache 2.0 — notable 92.1% HumanEval at 30B.
At a Glance
Where Sarvam 30B leads
- 92.1% HumanEval — excellent coding; competitive with models twice its size
- 80% MMLU-Pro — strong academic breadth
- Apache 2.0 — fully open
- Indian-origin open model: optimised for multilingual including Indic languages
Where it lags
- 49.0% Arena Hard v2 — moderate instruction quality
- Limited benchmark coverage (only 3 metrics)
Best for: Open-source coding with strong HumanEval; Indic language applications; Apache 2.0 Indian AI deployments.
What Sarvam 30B Is
Sarvam 30B is Sarvam AI's July 2025 large language model — a 30B model from Sarvam, an Indian AI company focused on Indic language capabilities. The 92.1% HumanEval is the standout score, making it one of the strongest coding models at the 30B scale from this period.
Specifications
| Field | Value |
|---|---|
| Organization | Sarvam AI |
| License | Apache 2.0 |
| HuggingFace | sarvamai/sarvam-30b |
| Release date | July 2025 |
| Parameters | 30B |
| Modality | Text only |
Pricing
Open weights under Apache 2.0 — self-host at no cost. Available via Sarvam API.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| HumanEval | 92.1% | Benchgen evaluation | 2025-07 |
| MMLU-Pro | 80% | Benchgen evaluation | 2025-07 |
| Arena Hard v2 | 49.0% | Benchgen evaluation | 2025-07 |
Sarvam 30B vs Alternatives
| Model | HumanEval | MMLU-Pro | Arena Hard | License |
|---|---|---|---|---|
| Sarvam 30B | 92.1% | 80% | 49.0% | Apache 2.0 |
| Qwen2.5 32B Instruct | 88.4% | — | — | Apache 2.0 |
| Mistral Small 3.1 24B | 88.4% | 66.8% | — | Apache 2.0 |
Sarvam 30B leads on HumanEval (92.1%) vs Qwen2.5 32B (88.4%). For Indic language support: Sarvam 30B is purpose-built.
Related Models & Benchmarks
- Compare: Qwen2.5 32B Instruct · Mistral Small 3.1 24B
- Benchmarks: HumanEval · MMLU-Pro · Arena Hard v2
Frequently Asked Questions
What is Sarvam 30B?
Sarvam AI's July 2025 Indian-origin 30B model scoring 92.1% HumanEval, 80% MMLU-Pro. Apache 2.0 — top HumanEval at 30B, optimised for Indic languages.Specs from Sarvam AI's Sarvam 30B release (July 2025) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.