# Qwen3 235B A22B (Instruct 2507)

> Qwen3-235B-A22B-Instruct-2507 is Qwen's updated non-thinking MoE model for multilingual instruction following, coding, tool use, and long-context work. It activates 22B of 235B parameters and provides a native 262K context window.

**ID**: `qwen/qwen3-235b-a22b`  
**Creator**: Qwen  
**Category**: text  
**Context**: 262K tokens  
**Pricing**: $0.09 / $0.35 per 1M · billed by your provider (BYOK)  
**Released**: 2025-07-21  
**Web page**: https://anyrouter.dev/model/qwen/qwen3-235b-a22b

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

**Tokenizer**: Qwen

**Capabilities**: chat, streaming, coding, function-calling

**Supported parameters**: max_tokens, temperature, top_p, stop, tools, tool_choice, response_format

**Aliases**: `qwen/qwen3-235b-a22b-2507`

## 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": "qwen/qwen3-235b-a22b",
  "messages": [
    {
      "role": "user",
      "content": "Say hi in 3 words."
    }
  ]
}'
```

## Providers

| Provider | Input | Output |
| --- | --- | --- |
| aihubmix-byok | $0.35/M (list) | $1.40/M (list) |
| openrouter-byok | $0.09/M (list) | $0.35/M (list) |

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**: 0.7
- **Aider Polyglot**: 0.6
- **MATH Level 5**: 0.7

Benchmark data from [Epoch AI](https://epoch.ai/benchmarks) (CC-BY 4.0).

## Model card

- Architecture: mixture of experts [1]
- Total parameters: 235B [1,2]
- Active parameters: 22B [1,2]
- Native context: 131072 tokens [1]
- License: Apache-2.0, open weights (https://huggingface.co/Qwen/Qwen3-235B-A22B) [1]

### Strengths [1,2]

- Thinking and non-thinking modes in one model
- Tool calling in both modes
- 119 languages and dialects

### Known limits [1]

- Context past 32,768 tokens needs YaRN, which can hurt short-text quality

### Use cases [1]

- Reasoning, math and coding in thinking mode
- Fast general chat in non-thinking mode
- Agents that call external tools

### API notes [1]

- enable_thinking switch, plus /think and /no_think soft switches
- Supports tool calling

### Links

- announcement: https://qwenlm.github.io/blog/qwen3/
- model card: https://huggingface.co/Qwen/Qwen3-235B-A22B
- paper: https://arxiv.org/abs/2505.09388
- docs: https://qwen.readthedocs.io/en/latest/

### Sources

1. [Qwen/Qwen3-235B-A22B model card](https://huggingface.co/Qwen/Qwen3-235B-A22B/raw/main/README.md) (accessed 2026-10-14)
2. [Qwen3: Think Deeper, Act Faster](https://qwenlm.github.io/blog/qwen3/) (accessed 2026-10-14)
3. [Qwen3 Technical Report](https://arxiv.org/abs/2505.09388) (accessed 2026-10-14)

## Source

- Upstream docs: https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507