Sarvam 105B
Sarvam 105B is a multilingual sparse language model trained from scratch in India. It emphasizes reasoning, coding, and agents across 22 Indian languages, using an MLA-style attention stack and context extension. Public access began through the limited Indus beta on February 20; Apache-licensed weight publication followed on March 6. This page represents the original open checkpoint, not later hosted conversational and work-agent specializations. The checkpoint supports up to 128K context, while reported evaluations use shorter task-specific limits. Model outputs require task-specific checks in languages and domains beyond the training focus.
2026-02-20
105B total, 10.3B active
Mixture-of-Experts Transformer
Apache-2.0
Specifications
- Parameters
- 105B total, 10.3B active
- Architecture
- Mixture-of-Experts Transformer
- License
- Apache-2.0
- Context Window
- 131,072 tokens
- Type
- text
- Modalities
- text
Benchmark Scores
Advanced Specifications
- Model Family
- Sarvam
- API Access
- Not Available
- Chat Interface
- Available
- Multilingual Support
- Yes
Capabilities & Limitations
- Capabilities
- multilingual generationreasoningcodingtool use
- Known Limitations
- Language performance varies by taskSparse activation does not remove total weight memory needsIndus beta access initially had capacity limits
- Notable Use Cases
- Indian-language assistantsprivate multilingual inferencecoding and agent research