# Sarvam 105B

> Sarvam's flagship 105B-parameter sparse Mixture-of-Experts model with 10.3B active parameters, built for reasoning, tool use and the ten most-spoken Indian languages. The 1711 weights run behind the stable sarvam-105b API id with a 128K context window.

**ID**: `sarvam/sarvam-105b`  
**Creator**: sarvam  
**Category**: text  
**Context**: 131K tokens  
**Pricing**: BYOK — billed by your provider  
**Released**: 2026-11-22  
**Links**: [Weights on Hugging Face](https://huggingface.co/sarvamai/sarvam-105b-1711) · [Announcement](https://www.sarvam.ai/blogs/sarvam-105b-1711) · [AnyRouter release](https://anyrouter.dev/blog/releases/sarvam-105b-1711)  
**Web page**: https://anyrouter.dev/model/sarvam/sarvam-105b

**Input modalities**: text  
**Output modalities**: text  

**Capabilities**: chat, streaming, function-calling, reasoning, agentic

**Aliases**: `sarvam/sarvam-105b-1711`

## Usage

**Endpoint**: `POST https://anyrouter.dev/api/v1/chat/completions`

```bash
curl https://anyrouter.dev/api/v1/chat/completions \
  -H "Authorization: Bearer $ANYROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "sarvam/sarvam-105b",
  "messages": [
    {
      "role": "user",
      "content": "Say hi in 3 words."
    }
  ]
}'
```

## Providers

| Provider | Input | Output |
| --- | --- | --- |
| sarvam-byok | BYOK — billed by your provider | — |
| sarvam-byok | BYOK — billed by your provider | — |

BYOK routes are billed by your own provider key — AnyRouter charges $0. Rates marked (list) are the provider's published list price.

## Benchmarks

- **GPQA Diamond**: 0.8

## Model card

- Architecture: mixture of experts [1,2]
- Total parameters: 105B [1,2]
- Active parameters: 10.3B [1]
- Native context: 131072 tokens [2]
- License: Apache-2.0, open weights (https://huggingface.co/sarvamai/sarvam-105b-1711) [1,2]

### Strengths [1,3]

- 105B MoE with 10.3B active parameters and multi-head latent attention.
- Post-trained for tool calling, instruction following and Indian languages.
- Apache 2.0 weights in NVFP4 and BF16.

### Known limits [2]

- Reasoning is on by default and reasoning tokens count toward max_tokens.
- The conversations API id has a smaller 32K context and never reasons.

### Use cases [3,2]

- Agentic tool use and long-context reasoning.
- Indic-language chat across native script, romanized and code-mixed input.

### API notes [2]

- OpenAI-compatible POST /v2/chat/completions; tools and tool_choice supported.
- Set reasoning_effort to none to turn thinking off on sarvam-105b.

### Links

- announcement: https://www.sarvam.ai/blogs/sarvam-105b-1711
- model card: https://huggingface.co/sarvamai/sarvam-105b-1711
- docs: https://docs.sarvam.ai/api/token-factory/models/sarvam-105b
- pricing: https://docs.sarvam.ai/api/getting-started/pricing

### Sources

1. [Sarvam-105B-1711 model card](https://huggingface.co/sarvamai/sarvam-105b-1711) (accessed 2026-11-23)
2. [Sarvam-105B API model page](https://docs.sarvam.ai/api/token-factory/models/sarvam-105b) (accessed 2026-11-23)
3. [Sarvam-105B-1711 launch post](https://www.sarvam.ai/blogs/sarvam-105b-1711) (accessed 2026-11-23)

## Source

- Upstream docs: https://docs.sarvam.ai/api/token-factory/models/sarvam-105b
