Qwen
Alibaba's open-weight family, with a coder model in nearly every size.
Qwen is Alibaba's model family and the largest open-weight line there is — dozens of sizes from a few billion parameters up to the Max models, most of them published so you can run them yourself. For vibe coders the draw is the Qwen Coder models: strong at code for the price, cheap to call through a provider, and small enough at the low end to run on your own machine. The flagship Max models take text, images and video and go up to a million tokens of context.
At a glance — Qwen3 Coder Next
- Cost
- $0.12 in · $0.80 out per 1M tokensBudget
- Memory
- 262K tokens — about 400 pages
- Reads
- Text
- Can do
- Use tools (agents)
- Structured output
- Thinking mode (not supported)
- Good at
- Coding: better than 34% · Agentic work: better than 8% · Overall: better than 8%
- Reliability
- 99.96% uptime, last 24h
- Released
- Feb 4, 2026
Advanced details
- API model ID
qwen/qwen3-coder-next- Max output
- 236K tokens
- Max input
- 205K tokens
- Cached input
- $0.07 per 1M
- Tokenizer
- Qwen
- Coding index
- 36.2
- Agentic work index
- 0.9
- Overall index
- 9.2
Providers
The same model, hosted by different companies. OpenRouter picks one per request and falls back to the next if it fails.
| Provider | Input /1M | Output /1M | Cached /1M | Max output | Uptime 24h |
|---|---|---|---|---|---|
| Parasail · bf16 | $0.12 | $0.80 | $0.07 | 236K | 99.96% |
| StreamLake | $0.18 | $0.90 | $0.036 | 64K | 99.52% |
| Novita · fp8 | $0.20 | $1.50 | — | 66K | 99.61% |
| Alibaba | $0.30 | $1.50 | — | 66K | 94.87% |
Supported parameters
- frequency_penalty
- logit_bias
- logprobs
- max_tokens
- presence_penalty
- repetition_penalty
- response_format
- seed
- stop
- structured_outputs
- temperature
- tool_choice
- tools
- top_k
- top_logprobs
- top_p
Live from OpenRouter, refreshed hourly. Scores from Artificial Analysis.
Key info
- Pricing
- Open source
- Category
- Models
- Best for
- Intermediate
- Made by
- Alibaba
- Weights
- Open, most sizes
- Good for
- Cheap coding models, and running one yourself