Alibaba Qwen's next-generation code model. Optimized for repository-level understanding and multi-turn tool interaction, at performance close to Qwen3-Coder-Plus.
Qwen3-Coder-Next is Alibaba Cloud Model Studio's balanced managed Qwen-Coder option for everyday agentic coding. In TheRouter it is exposed as a text-in, text-out OpenAI-compatible chat model with streaming, tool calls, structured JSON, a 262,144-token catalog context window, and 65,536-token maximum output.
Use it when teams want a practical default between Plus and Flash: repository search, medium patches, generated tests, code review triage, CI-failure explanations, and IDE assistant workloads where quality, speed, and cost all matter. Alibaba's Qwen-Coder material positions this family around coding agents, function calling, repository context, and external-environment interaction.
The caveat is evidence discipline. Public sources support Qwen3-Coder family capabilities and Alibaba's model-list/pricing facts, but they do not publish a complete Next-specific benchmark table. This page therefore treats Next as the balanced managed coder, not as a benchmark-proven quality ceiling; escalate high-risk autonomous edits to Plus, a larger Qwen-Coder sibling, or another frontier coder after your own evals.
Alibaba Cloud operates Qwen3-Coder-Next as a DashScope / Model Studio managed model with its own compatible-mode endpoints, regional routing, and multi-band input-length pricing.
TheRouter exposes it as qwen/qwen3-coder-next on the shared OpenAI-compatible /v1/chat/completions endpoint. The public catalog currently stores the base rate and the 32K-128K long-context tier; Alibaba's highest 128K-256K input band is noted in source comments but cannot be represented by the single-threshold schema, so very large prompts should be cost-tested before production rollout.
| Type | Rate |
|---|---|
| Input | $0.324 / 1M tokens |
| Output | $1.62 / 1M tokens |
| Managed SKU | qwen3-coder-nextwww.alibabacloud.com ↗ | verified |
| TheRouter public id | qwen/qwen3-coder-nextwww.alibabacloud.com ↗ | verified |
| Context window | 262,144 tokens in TheRouter catalogwww.alibabacloud.com ↗ | verified |
| Max output | 65,536 tokens in TheRouter catalogwww.alibabacloud.com ↗ | verified |
| Modalities | Text input, text outputwww.alibabacloud.com ↗ | verified |
| Supported operations | Chat completions, streaming, tool calls, tool_choice, response_format, stop sequenceswww.alibabacloud.com ↗ | verified |
| Pricing structure | Tiered by input length; TheRouter currently stores the base tier plus the 32K-128K long-context tierwww.alibabacloud.com ↗ | verified |
| Training cutoff | Not publicly disclosed | unknown |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Qwen3-Coder family agentic coding Qwen's launch material reports open-model SOTA agentic coding, browser-use, and tool-use performance for the Qwen3-Coder family / 480B flagship. This supports the family positioning, not a Next-specific numeric claim. | Family-level evidence; Next-specific public score not disclosed | qwenlm.github.io ↗ | |
Tool calling / external environment workflows Alibaba's Qwen-Coder documentation covers tool-calling and external-environment interaction patterns. Use this as capability evidence and run project-specific evals for exact reliability. | Documented capability; no independent Next score published | alibabacloud.com ↗ | |
Public Next-specific benchmark disclosure Alibaba publishes model positioning, API behavior, and pricing for the managed Coder SKUs, but this pass found no official SWE-bench, LiveCodeBench, HumanEval, or Terminal-Bench table specifically for qwen3-coder-next. | Not publicly disclosed | alibabacloud.com ↗ |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "qwen/qwen3-coder-next",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Use the standard OpenAI-compatible chat endpoint for code edits, explanations, and review triage.
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen/qwen3-coder-next",
"messages": [
{"role": "system", "content": "You are a senior software engineer. Return concise patches and tests."},
{"role": "user", "content": "Explain this failing test and suggest the smallest fix: ..."}
],
"temperature": 0.3,
"max_tokens": 4096
}'Alibaba Cloud's Model Studio guide lists qwen3-coder-next among Qwen models callable through its OpenAI-compatible Responses API, alongside chat-completions compatibility for teams migrating agent workloads.
The Qwen3-Coder launch framed the family around agentic coding, repository-scale context, tool use, Qwen Code, and OpenAI-compatible integration paths for developer tools.
Choose Next when the workload is coding-agent shaped and the team wants a balanced default: repository search, generated tests, medium patches, CI explanation, and code-review triage. Escalate to Plus, 480B, or a frontier coder when the change is autonomous, high-risk, or hard to evaluate automatically; use Flash when cost and throughput dominate quality.
Yes. TheRouter's catalog lists tools and tool_choice for qwen/qwen3-coder-next, and Alibaba's Qwen-Coder documentation describes function-calling patterns for coding-agent workflows.
No. Alibaba documents input-length pricing bands for Qwen-Coder models, and TheRouter's current catalog can represent only a base tier plus one long-context tier. Validate cost on realistic prompt sizes before allowing agents to send very large repositories.
Yes. TheRouter exposes qwen/qwen3-coder-next through the OpenAI-compatible chat endpoint. Set the client's base URL to https://api.therouter.ai/v1, use your TheRouter API key, and set the model name exactly to qwen/qwen3-coder-next.
Not as a complete official Next-specific table in the public sources checked for this pass. The page therefore uses high-confidence sourcing for capabilities and pricing, and it labels benchmark evidence as family-level or not publicly disclosed rather than inventing scores.
| source | URL | retrieved | |
|---|---|---|---|
| Managed SKU | www.alibabacloud.com ↗ | 2026-08-18 | verified |
| TheRouter public id | www.alibabacloud.com ↗ | 2026-08-18 | verified |
| Context window | www.alibabacloud.com ↗ | 2026-08-18 | verified |
| Max output | www.alibabacloud.com ↗ | 2026-08-18 | verified |
| Modalities | www.alibabacloud.com ↗ | 2026-08-18 | verified |
| Supported operations | www.alibabacloud.com ↗ | 2026-08-18 | verified |
| Pricing structure | www.alibabacloud.com ↗ | 2026-08-18 | verified |
| Training cutoff | — | — | unknown |
| Qwen3-Coder family agentic coding | qwenlm.github.io ↗ | 2026-08-18 | verified |
| Tool calling / external environment workflows | alibabacloud.com ↗ | 2026-08-18 | verified |
| Public Next-specific benchmark disclosure | alibabacloud.com ↗ | 2026-08-18 | verified |
| DashScope documents OpenAI Responses API support for Qwen models | help.aliyun.com ↗ | 2026-08-18 | verified |
| Qwen launches the Qwen3-Coder family and Qwen Code | qwenlm.github.io ↗ | 2026-08-18 | verified |
| When should I choose qwen/qwen3-coder-next instead of Plus or Flash? | www.alibabacloud.com ↗ | 2026-08-18 | to verify |
| Does Qwen3-Coder-Next support tool calling? | www.alibabacloud.com ↗ | 2026-08-18 | to verify |
| Is the 1M context free to use at the same price as small prompts? | www.alibabacloud.com ↗ | 2026-08-18 | to verify |
| Can I use it from Cursor, Cline, or another OpenAI-compatible client? | www.alibabacloud.com ↗ | 2026-08-18 | to verify |
| Are Next benchmark scores published? | www.alibabacloud.com ↗ | 2026-08-18 | to verify |