Alibaba Qwen's low-cost coding agent model. Inherits Qwen3-Coder-Plus agentic tool use with 1M context and improved tool-calling stability.
Qwen3-Coder-Flash is Alibaba Cloud Model Studio's low-cost managed member of the Qwen-Coder line. In TheRouter it is the budget and throughput-oriented coding-agent option: text in, text out, OpenAI-compatible chat, streaming, tool calls, structured JSON, and a 1,000,000-token operational context window from the active catalog entry. Since the Oct 15, 2026 update of Alibaba's text-generation guide, Alibaba lists qwen3-coder-flash (snapshot qwen3-coder-flash-2025-07-28) under legacy models that are no longer recommended. It is still served and remains available on TheRouter; for coding tools, Alibaba now recommends qwen3.7-plus (qwen/qwen3.7-plus).
Use it when the job is still agentic coding, but the economics matter more than peak reasoning: repository search, small-to-medium patches, generated tests, code review triage, CI-failure explanations, and high-QPS developer-tool backends. Alibaba positions Qwen-Coder models around coding agents, function calling, and external-environment interaction; Flash is the cost-sensitive SKU in that managed family.
The important caveat is that public benchmark evidence is mostly family-level rather than Flash-specific. Treat Qwen3-Coder-Flash as the economical operational choice inside a Qwen-Coder routing policy, not as the proof-backed quality ceiling; route higher-stakes autonomous edits to qwen/qwen3.7-plus (Alibaba's current coding-tools recommendation), qwen/qwen3.8-max for the strongest reasoning, qwen/qwen3-coder-480b, or another frontier coding model after your own evals. When the goal is only lower cost, Alibaba points coding-tool users to qwen3.8-flash (qwen/qwen3.8-flash).
Alibaba Cloud operates Qwen3-Coder-Flash 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-flash on the shared OpenAI-compatible /v1/chat/completions endpoint. The public catalog currently represents Alibaba's multiple price bands as a base tier plus one long-context tier, so very large prompts should be cost-tested before production rollout.
| Type | Rate |
|---|---|
| Prompts up to 32K tokens | |
| Input | $0.324 / 1M tokens |
| Output | $1.62 / 1M tokens |
| Cached input | $0.324 / 1M tokens |
| Prompts over 32K tokens | |
| Input | $0.540 / 1M tokens |
| Output | $2.70 / 1M tokens |
| Cached input | $0.540 / 1M tokens |
Tiered pricing: when a request's prompt (input, cached tokens included) is over 32K tokens, every token in that request is billed at the long-context rates; exactly 32K is still the base rate. This model publishes no separate cached-input price: cached tokens are billed at the input rate of their group (the "Cached input" row shows that rate).
| Managed SKU | qwen3-coder-flashwww.alibabacloud.com ↗ | verified |
| TheRouter public id | qwen/qwen3-coder-flashwww.alibabacloud.com ↗ | verified |
| Context window | 1,000,000 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 one base tier and one 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 Flash-specific numeric claim. | Family-level evidence; Flash-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 Flash score published | alibabacloud.com ↗ | |
Public Flash-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-flash. | 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-flash",
"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-flash",
"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
}'The Oct 15, 2026 update of the Model Studio text-generation guide lists qwen3-coder-flash as legacy and no longer recommended. For coding tools, Alibaba now recommends qwen3.7-plus, with qwen3.8-max for the strongest reasoning and qwen3.8-flash to cut costs. qwen/qwen3-coder-flash stays available on TheRouter.
Alibaba Cloud's Model Studio guide lists qwen3-coder-flash 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.
No longer recommended upstream. Since the Oct 15, 2026 docs update, Alibaba lists qwen3-coder-flash under legacy models; it is still served and stays available on TheRouter, so existing integrations keep working. For coding tools Alibaba recommends qwen/qwen3.7-plus, qwen/qwen3.8-max for the strongest reasoning, and qwen/qwen3.8-flash when cost matters most. Run your own evals before moving triage, test-generation, or CI-explanation traffic.
Yes. TheRouter's catalog lists tools and tool_choice for qwen/qwen3-coder-flash, 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-flash 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-flash.
Not as a complete official Flash-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 ↗ | 2026-08-17 | verified |
| TheRouter public id | www.alibabacloud.com ↗ | 2026-08-17 | verified |
| Context window | www.alibabacloud.com ↗ | 2026-08-17 | verified |
| Max output | www.alibabacloud.com ↗ | 2026-08-17 | verified |
| Modalities | www.alibabacloud.com ↗ | 2026-08-17 | verified |
| Supported operations | www.alibabacloud.com ↗ | 2026-08-17 | verified |
| Pricing structure | www.alibabacloud.com ↗ | 2026-08-17 | verified |
| Training cutoff | — | — | unknown |
| Qwen3-Coder family agentic coding | qwenlm.github.io ↗ | 2026-08-17 | verified |
| Tool calling / external environment workflows | alibabacloud.com ↗ | 2026-08-17 | verified |
| Public Flash-specific benchmark disclosure | alibabacloud.com ↗ | 2026-08-17 | verified |
| Alibaba moves Qwen3-Coder-Flash to its legacy model list | alibabacloud.com ↗ | 2026-10-19 | verified |
| DashScope documents OpenAI Responses API support for Qwen models | help.aliyun.com ↗ | 2026-08-17 | verified |
| Qwen launches the Qwen3-Coder family and Qwen Code | qwenlm.github.io ↗ | 2026-08-17 | verified |
| Is qwen/qwen3-coder-flash still recommended, and what should I use instead? | alibabacloud.com ↗ | 2026-10-19 | to verify |
| Does Qwen3-Coder-Flash support tool calling? | www.alibabacloud.com ↗ | 2026-08-17 | to verify |
| Is the 1M context free to use at the same price as small prompts? | www.alibabacloud.com ↗ | 2026-08-17 | to verify |
| Can I use it from Cursor, Cline, or another OpenAI-compatible client? | www.alibabacloud.com ↗ | 2026-08-17 | to verify |
| Are Flash benchmark scores published? | www.alibabacloud.com ↗ | 2026-08-17 | to verify |