10M-token long-context Qwen — built for document-scale analysis.
Alibaba Bailian documents qwen-long as a CN-hosted long-context model with 10M-token context, 8,192-token output, text-in/text-out, direct HTTP guidance of 1M tokens, and file submission recommended above that length.
TheRouter exposes qwen/qwen-long on the OpenAI-compatible chat path with a catalog context length of 10,485,760 tokens and CN-region Bailian serving. The page therefore treats 10M as the vendor contract and 10,485,760 as TheRouter's operational catalog value, and warns that direct long prompts need validation.
qwen-long is the long-context specialist of the Qwen family. With a 10M-token context window it ingests entire books, multi-hundred-page legal filings, whole codebases, or weeks of chat history in a single request, then answers, summarises, or extracts across the full input.
This model is deployed only on the mainland Bailian endpoint (bailian-cn). It does not have a Singapore failover — if bailian-cn is unavailable the model is unavailable. Plan for that in your retry strategy.
| Type | Rate |
|---|---|
| Input | $0.1296 / 1M tokens |
| Output | $0.486 / 1M tokens |
| Context length | 10,000,000 tokens documented by Alibaba; TheRouter catalog exposes 10,485,760 tokenshelp.aliyun.com ↗ | verified |
| Direct HTTP input limit | 1,000,000 tokens; Alibaba recommends file submission above this lengthhelp.aliyun.com ↗ | verified |
| Maximum output | 8,192 tokenshelp.aliyun.com ↗ | verified |
| Modalities | Text input → text outputhelp.aliyun.com ↗ | verified |
| Unsupported upstream features | Function calling, structured output, web search, prefix completion, context cache, batch inference, and fine-tuninghelp.aliyun.com ↗ | verified |
| Published mainland China quota | qwen-long: 1,200 RPM and 3,000,000 TPM in cn-beijinghelp.aliyun.com ↗ | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Public long-context benchmark suite Alibaba's qwen-long page describes target tasks and limits but does not publish a benchmark table for this model. | — | Not publicly disclosed | — |
Document QA benchmark Use your own long-document acceptance set before replacing a chunked retrieval pipeline. | — | Not publicly disclosed | — |
Latency / throughput benchmark The model is designed for very large contexts; measure TTFT and completion latency at your intended prompt size. | — | Not publicly disclosed | — |
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/qwen-long",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Use the OpenAI-compatible chat endpoint for long text prompts. Keep direct HTTP requests under the upstream 1M-token guidance unless your integration has an approved file-submission path.
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"qwen/qwen-long","messages":[{"role":"user","content":"Summarize the attached long document text: ..."}],"max_tokens":1200}'The qwen-long documentation lists the stable qwen-long route, a dynamically updated qwen-long-latest, and the snapshot qwen-long-2025-01-25. All three carry the same 10M context and 8,192-token output headline, while the documented quota differs between stable/latest and the snapshot route.
Alibaba documents qwen-long with 10,000,000-token maximum input and context length. The same page also says direct HTTP submission supports 1M tokens and recommends file submission above that length. TheRouter's catalog exposes 10,485,760 tokens, so treat 10M as the vendor-published contract and validate the exact request shape you plan to run.
Alibaba's qwen-long capability table says function calling and structured output are not supported, along with web search, prefix completion, context cache, batch inference, and model fine-tuning. Through TheRouter, use plain chat prompts and treat JSON output as best-effort instruction following.
Choose qwen-long only when context size is the constraint: whole-document review, very long transcript Q&A, cross-section extraction, or archive-level summarisation. For normal chat, short classification, or latency-sensitive flows, qwen-plus will usually be simpler and faster.
| source | URL | retrieved | |
|---|---|---|---|
| Context length | help.aliyun.com ↗ | 2026-08-01 | verified |
| Direct HTTP input limit | help.aliyun.com ↗ | 2026-08-01 | verified |
| Maximum output | help.aliyun.com ↗ | 2026-08-01 | verified |
| Modalities | help.aliyun.com ↗ | 2026-08-01 | verified |
| Unsupported upstream features | help.aliyun.com ↗ | 2026-08-01 | verified |
| Published mainland China quota | help.aliyun.com ↗ | 2026-08-01 | verified |
| Public long-context benchmark suite | help.aliyun.com ↗ | 2026-08-01 | unknown |
| Document QA benchmark | help.aliyun.com ↗ | 2026-08-01 | unknown |
| Latency / throughput benchmark | help.aliyun.com ↗ | 2026-08-01 | unknown |
| Alibaba lists qwen-long snapshot and qwen-long-latest alongside the base route | help.aliyun.com ↗ | 2026-08-01 | verified |
| Is qwen/qwen-long really a 10M-token model? | help.aliyun.com ↗ | 2026-08-01 | to verify |
| Does qwen-long support function calling or structured output? | help.aliyun.com ↗ | 2026-08-01 | to verify |
| When should I choose qwen-long instead of qwen-plus? | help.aliyun.com ↗ | 2026-08-01 | to verify |