What is Qwen3.7-Max?
Qwen3.7-Max is the largest model in Alibaba's Qwen3.7 series, sold only through the Alibaba Cloud Model Studio API; Alibaba never released its weights. It handles up to 1M tokens per request in thinking or non-thinking mode. The default model ID reads text only, and a later snapshot adds image input. Alibaba built it for coding agents, office automation and runs that last for hours.
It is now a legacy model: Alibaba no longer recommends it, Qwen Chat does not offer it, and Qwen3.8-Max scores higher on Alibaba's own tests at the same or a lower price, with open weights.
Is Qwen3.7-Max still available?
Yes, on the Model Studio API at unchanged prices, but Alibaba lists it as a legacy model it no longer recommends, and Qwen Chat does not offer it.
- API
- Still callable as qwen3.7-max in the Model Studio regions that list it, from $1.65 per million input tokens.
- Qwen Chat
- Offers Qwen3.8-Max and Qwen3.7-Plus instead.
- Recommended model
- Alibaba points developers who want the strongest reasoning to Qwen3.8-Max.
- Retirement notice
- Alibaba promises three months' notice before it retires a mainline model like this one.
Who is Qwen3.7-Max best for?
Qwen3.7-Max is best for workloads that must run inside the US, teams whose agents are already tuned to it, and long coding or office agent runs in tools such as Claude Code and OpenClaw.
Workloads that must stay in the US
Its US deployment scope keeps inference inside the US (Virginia) region, and no other Qwen Max model is offered that way on Model Studio.
Pipelines already built on it
Teams with prompts and tests tuned to Qwen3.7-Max can keep calling it at the same price while they try Qwen3.8-Max.
Long agent runs
Alibaba shows a 35-hour unattended kernel tuning run with 1,158 tool calls that ended with a kernel 10 times faster than the reference.
Coding in Claude Code or OpenClaw
The API speaks both the OpenAI and Anthropic formats, so these tools switch to it with a new base URL, model name and key.
What's included with Qwen3.7-Max?
Thinking you can switch off, OpenAI- and Anthropic-compatible APIs, cheap cache hits, tool calling with structured output, and a choice of where inference runs.
- Thinking on or off
- Both modes cost the same, and a preserve_thinking option keeps earlier reasoning in context for agent tasks.
- OpenAI and Anthropic formats
- Chat completions and responses APIs in OpenAI's format, plus an Anthropic-compatible endpoint for Claude Code.
- Context caching
- Cache hits are billed at 10% of the input price, which cuts the cost of long agent loops.
- Tools and structured output
- Function calling, Model Studio's built-in tools and structured output are all supported.
- Choice of deployment scope
- Run inference worldwide on the Global scope, inside the US, or from Singapore; stored data stays in the region you pick.
How much does Qwen3.7-Max cost?
Qwen3.7-Max is billed per token on Alibaba Cloud Model Studio, from $1.65 per million input tokens and $4.951 per million output tokens on the Global scope, or $2.50 and $7.50 on the US and Singapore scopes; there is no subscription.
- Offered in the US (Virginia), Hong Kong, Frankfurt and Tokyo regions
- Inference can run in any region, for a larger pool of capacity
- Up to 1M input tokens per request at one price
- US (Virginia) region, with inference kept inside the US
- Thinking and non-thinking modes cost the same
- 1M free tokens, valid for 90 days
- The only region with a free quota
These are pay-as-you-go API prices on Alibaba Cloud Model Studio; there is no subscription, and Alibaba's Coding Plan does not include Qwen3.7-Max. In the China (Beijing) region it costs $1.65 / $4.951 per million input / output tokens, and batch calls there cost half. Cache hits are billed at 10% of the standard input price and creating an explicit cache at 125%; caching and batch discounts do not stack. The Singapore free quota runs 90 days from Model Studio activation or the model's release, whichever is later. Full pricing on alibabacloud.com. We checked these prices on .
How do readers rate Qwen3.7-Max?
No one has rated Qwen3.7-Max yet. Be the first.
Sign in to rate. One rating per account.
Qwen3.7-Max pros and cons
It scores well on Alibaba's own coding and agent tests and suits long runs, but its weights are closed, it has no flat-rate plan, and Alibaba now lists it as legacy.
Pros
- In Alibaba's own table it beats Claude Opus 4.6 on Terminal Bench 2.0 (69.7 against 65.4) and SWE-Pro (60.6 against 57.3).
- Built for long runs: Alibaba's 35-hour unattended kernel job made 1,158 tool calls and still found gains after 30 hours.
- Early users say it handles long agent loops well while spending fewer reasoning tokens than Claude Opus, which keeps bills down.
- It is the only Qwen Max model Alibaba offers with a US-only deployment scope.
Cons
- Closed weights: you can only call it through Alibaba's API, a point many Hacker News and Reddit users raised at launch.
- Now legacy: Alibaba no longer recommends it, Qwen Chat does not offer it, and Qwen3.8-Max scores higher at the same or a lower price.
- No flat-rate plan: Alibaba's Coding Plan does not cover it, and Reddit users hunting for a subscription to use it in coding tools came up short.
- Some companies' security rules bar sending data to a Chinese cloud provider, a concern several Hacker News commenters raised.
What are the best Qwen3.7-Max alternatives?
GLM-5.3 for a flat monthly plan and open weights, Claude Opus 5.5 for a Claude subscription, and Qwen3.6-27B to run an Alibaba model yourself.
GLM-5.3FreeChoose it when you want a flat monthly price or weights you can host; Z.ai's Coding Plan starts at $18 a month, and its API costs $1.4 / $4.4 per million tokens with a 1M context.
Claude Opus 5.5PaidChoose it when you want a current frontier model inside Claude Code on a subscription; it is on Claude Pro from $17 a month, or $4 / $20 per million tokens on the API.
How does Qwen3.7-Max compare with Qwen3.8-Max?
Qwen3.8-Max scores higher across Alibaba's comparison table, costs the same or less, and has open weights; Qwen3.7-Max keeps only its US-only scope.
- Coding
- Terminal Bench 2.1 rises from 74.5 to 86.6 and SWE-bench Pro from 60.6 to 67.7 in Alibaba's table.
- Agents
- Toolathlon Verified rises from 49.7 to 72.5.
- Price
- Both cost $1.65 / $4.951 per million tokens on the Global scope; in Singapore Qwen3.8-Max costs $2 / $6 against $2.50 / $7.50.
- Weights
- Qwen3.8-Max, a 2.4T-parameter model with 95B active, has open weights on Hugging Face; Qwen3.7-Max never did.
Questions about Qwen3.7-Max
What is Qwen3.7-Max?
Qwen3.7-Max is Alibaba's closed, API-only model for coding agents, office work and long autonomous runs, with a 1M-token context; Alibaba now lists it as legacy behind Qwen3.8-Max.
Who is Qwen3.7-Max for?
Qwen3.7-Max is best for workloads that must run inside the US, teams whose agents are already tuned to it, and long coding or office agent runs in tools such as Claude Code and OpenClaw.
Is Qwen3.7-Max free?
Mostly no. Model Studio accounts in the Singapore region get 1 million free tokens for Qwen3.7-Max, valid for 90 days; after that, and in every other region, each call is billed per token, from $1.65 per million input tokens.
How much does Qwen3.7-Max cost?
Qwen3.7-Max's paid plans are Global deployment at $1.65 / $4.951 per million input / output tokens, US deployment at $2.5 / $7.5 per million input / output tokens and International (Singapore) at $2.5 / $7.5 per million input / output tokens.
What are the best Qwen3.7-Max alternatives?
The closest alternatives on Recaply are GLM-5.3 and Claude Opus 5.5.
Where does Qwen3.7-Max process my data?
It depends on the deployment scope you pick. The US scope keeps inference inside the US, the Global scope can run it in any region for more capacity, and the Singapore region uses its International scope. Stored data stays in the region you chose.
Is Qwen3.7-Max open source?
No. Alibaba calls Qwen3.7-Max a proprietary model and never released its weights, so you can only call it through the Model Studio API. Its successor, Qwen3.8-Max, is the first Qwen Max model with open weights.
Can I use Qwen3.7-Max in Claude Code?
Yes. Model Studio accepts the Anthropic API format, so you point Claude Code at Alibaba's endpoint, set the model to qwen3.7-max and use a Model Studio API key. It also works in OpenAI-compatible tools such as OpenClaw and Qwen Code.
Does Alibaba's Coding Plan include Qwen3.7-Max?
No. The Coding Plan covers only a fixed list of models, including Qwen3.7-Plus, Qwen3.6-Plus, Kimi K2.5, GLM-5 and MiniMax-M2.5, but not Qwen3.7-Max. You pay for Qwen3.7-Max per token through Model Studio.
Our verdict
Qwen3.7-Max was a strong agent model, but new projects should start on Qwen3.8-Max.
Alibaba now lists it as legacy, and Qwen3.8-Max beats it on Alibaba's own coding and agent tests at the same or a lower price, with open weights.
- Skip it for
- New projects, and anyone who wants weights they can host or a flat monthly plan.
- Check first
- If your data must stay in the US, compare scopes first, since Qwen3.8-Max has no US-only scope on Model Studio.
Qwen3.8-Flash
Qwen3.6-Plus
Qwen3.6-Max-Preview
Qwen3.7-Plus
Qwen3-Max-Thinking
Qwen3.8-Max