Leaderboard / Sarvam AI
Sarvam 30B
Sarvam AI · Sarvam · released 2026-03-03 · data checked 2026-11-01 Needs verification
Sarvam 30B is a 32.2B-parameter Mixture-of-Experts model (2.4B active per token) from Sarvam AI (Rest of Asia), released 2026-03-03 under the Apache-2.0 license. It scores 61.6 on the Small LM Index, #18 of 34 medium models. It needs about 20.8 GB of memory at 4-bit with an 8K-token context.
Trained from scratch in India with a focus on Indian languages. Active parameter count is an estimate.
Benchmarks
Each bar shows this model's score. The tick marks the best medium model on that benchmark. Sources: 7 reported by Sarvam AI. How we label sources.
Not reported: estimated
What the index model expects from this model's other scores. Not a measurement.
Specifications
- Lab
- Sarvam AI (IN · Rest of Asia)
- Architecture
- Mixture-of-Experts
- Total parameters
- 32.2B
- Active parameters
- 2.4B
- Size class
- Medium
- Context window
- 131K tokens
- Input modalities
- text
- Reasoning
- Reasoning (thinking)
- License
- Apache-2.0
- Openness
- Open weights
- Memory to run
- ~20.8 GB at 4-bit, 8K context (more sizes)
- Hugging Face
- 5K downloads (30 days) · 234 likes
Compare with
- Sarvam 30B vs Nanbeige4.3 33B (33.1B, index 77.8)
- Sarvam 30B vs Nemotron 3.5 Lightning (31.6B, index 63.6)
- Sarvam 30B vs Nemotron 3 Nano 30B-A3B (31.6B, index 62.6)
- Sarvam 30B vs EXAONE 4.0 32B (32B, index 62.0)
- Sarvam 30B vs OLMo 3.1 32B Think (32.2B, index 57.0)
- Sarvam 30B vs Qwen3 32B (32.8B, index 52.4)
Sources
Something wrong here? Report it.
Run it locally
No GGUF build from a well-known publisher yet.
vllm serve sarvamai/sarvam-30bMemory needed: weights, context cache and runtime buffers.
| Quantization | 4K context | 32K context | 131K context |
|---|---|---|---|
| 16-bit F16 | 66 GB | 66 GB | 68 GB |
| 8-bit Q8_0 | 35 GB | 36 GB | 38 GB |
| 4-bit Q4_K_M | 21 GB | 21 GB | 23 GB |