# Qwen 3 Coder

> Purpose-built for code generation.

Human version: https://kymaapi.com/models/qwen-3-coder
Live JSON: `GET https://kymaapi.com/v1/models` (no auth required)

## Facts

- **Model ID**: `qwen-3-coder` — pass this as `model` in the request body
- **Creator**: Alibaba
- **Released**: 2025-07-22
- **Context window**: 131K tokens
- **Max output**: 33K tokens per response — a hard ceiling, not a default
- **Price**: $0.297 in / $1.35 out per 1M, cached input 10%
- **Capabilities**: tools, caching
- **Alias**: `code` resolves to this model

## Call it

```bash
curl https://kymaapi.com/v1/chat/completions \
  -H "Authorization: Bearer $KYMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "qwen-3-coder", "messages": [{"role": "user", "content": "Hello"}]}'
```

## Positioning

Qwen 3 Coder is a text-only language model optimized for code generation and debugging. Developers building code assistants, automated debugging workflows, or agentic systems should use it when they need strong reasoning and tool-use capabilities at a balanced cost.

## About

Created by Alibaba, this model is purpose-built for code generation and debugging. It supports tool use, structured outputs, and reasoning, with a 131,072-token context window and a 32,768-token maximum output. It processes text-only inputs and outputs.

On Kyma, it routes through an OpenAI-compatible endpoint with automatic failover and a 100% observed success rate. It runs at a median speed of 92 tokens per second. Prompt caching is supported, reducing repeated prefix costs by 90%, and the exact request cost is returned in the usage.cost field.

It does not support vision or multimodal inputs. The medium speed tier means it is not optimized for ultra-low-latency streaming. Output is capped at 32,768 tokens, which may be restrictive for generating extremely large codebases in a single pass.

## Use cases

- **Automated Code Generation** — Translates plain-text requirements into functional code snippets.
- **Interactive Code Debugging** — Identifies logical errors and suggests fixes from stack traces.
- **Agentic Development Workflows** — Coordinates tool calls and structured outputs for multi-step coding tasks.
- **Conversational Coding Sessions** — Maintains context across long debugging sessions within a 131k token window.

## Not ideal for

Avoid this model when you need image analysis, multimodal inputs, or ultra-low latency responses for real-time applications.

## Pick something else when

- You need vision or multimodal inputs → `gemma-4-31b`
- You need faster inference speeds → `deepseek-v4-flash`, `step-3.7-flash`
- You need lower-cost general chat → `llama-3.3-70b`

## See also

- All models: https://kymaapi.com/models.md
- Pricing: https://kymaapi.com/pricing.md
- Other models by Alibaba: https://kymaapi.com/models?q=Alibaba