Qwen 3.7 Models Compared: Max vs Plus vs Flash Pricing and API (2026)

Qwen 3.7 is Alibaba's proprietary model series. qwen3.7-max costs $2.50/$7.50 per 1M tokens, qwen3.7-plus $0.40/$1.60 up to 256K, and qwen3.7-flash $0.03/$0.13 up to 32K, all with a 1M context. Model IDs, snapshots, context tiers, benchmarks, and where to call it.

November 18, 2026 · 1 min read
Qwen 3.7 Models Compared: Max vs Plus vs Flash Pricing and API (2026)

Qwen 3.7 is Alibaba's proprietary model series, served through Alibaba Cloud Model Studio. It has three tiers. qwen3.7-max is the text-only flagship, first released May 20, 2026, at $2.50 input and $7.50 output per million tokens. qwen3.7-plus, snapshot May 26, 2026, takes text, images, and video and costs $0.40 and $1.60 up to 256K input tokens. qwen3.7-flash, snapshot July 15, 2026, is the cheap vision-language tier at $0.03 and $0.13 up to 32K. All three accept 1M tokens of context and support thinking mode, function calling, and structured output. The weights are not public, so you cannot self-host Qwen 3.7. Alibaba now lists qwen4-max ($1.50/$4.50) for the hardest tasks and still recommends qwen3.7-plus as the default starting model. OpenRouter resells all three tiers, and Fireworks serves Qwen 3.7 Plus on its serverless API.

1M
Context window, all three tiers
$2.50 / $7.50
qwen3.7-max per 1M tokens
$0.40 / $1.60
qwen3.7-plus per 1M, up to 256K
$0.03 / $0.13
qwen3.7-flash per 1M, up to 32K

What Is Qwen 3.7?

Qwen 3.7 is the generation of Alibaba's closed Qwen models that sits between Qwen 3.6 and Qwen 3.8. Alibaba ships it in three sizes under one family name. You reach it through Alibaba Cloud Model Studio, the platform formerly called DashScope, or through resellers such as OpenRouter.

Max came first. Alibaba published a thinking-only preview snapshot dated May 17, 2026, then the stable qwen3.7-max-2026-05-20 with thinking and non-thinking modes. Plus followed with the May 26 snapshot and Flash with the July 15 snapshot. By November 2026, Alibaba's model guide sends the hardest reasoning work to qwen4-max and keeps qwen3.7-plus as the model to start with.

Qwen 3.7 Max vs Plus vs Flash

Qwen 3.7 tiers on Alibaba Cloud Model Studio (International), checked November 18, 2026
TierModel IDInput typesContextPrice per 1M (in / out)
Maxqwen3.7-maxText1M$2.50 / $7.50
Plusqwen3.7-plusText, image, video1M (128K max output)$0.40 / $1.60 up to 256K
Flashqwen3.7-flashText, image, video1M$0.03 / $0.13 up to 32K

All three tiers support thinking mode, function calling, built-in tools, and structured output on Model Studio. Max is the only text-only tier. Plus is the general-purpose agent model, and Flash is the one to use for high-volume classification, extraction, and cheap subagent turns.

Qwen 3.7 Pricing by Context Tier

Plus and Flash bill by how many input tokens a request carries. A request with a 300K-token prompt pays the 256K-to-1M rate on every token. Max has one flat rate up to 1M.

Qwen 3.7 per-request input tiers (USD per 1M tokens, International)
ModelInput tokens per requestInputOutput
qwen3.7-max0 to 1M$2.50$7.50
qwen3.7-plus0 to 256K$0.40$1.60
qwen3.7-plus256K to 1M$1.20$4.80
qwen3.7-flash0 to 32K$0.03$0.13
qwen3.7-flash32K to 256K$0.10$0.40
qwen3.7-flash256K to 1M$0.20$0.80
Discounts that change the bill

Alibaba lists a limited-time 20% discount on qwen3.7-plus, which is why OpenRouter shows it at $0.32/$1.28. Flash batch calls are half price. All three tiers discount context-cache hits, which matters for agents that re-send the same long prefix on every turn. See prompt caching for how that math works.

Qwen 3.7 API: Model IDs and Snapshots

The unversioned IDs move when Alibaba ships a new snapshot. Pin a dated ID if you need repeatable output.

Qwen 3.7 model IDs on Model Studio
Model IDPoints toModes
qwen3.7-maxqwen3.7-max-2026-05-20Thinking and non-thinking
qwen3.7-max-2026-06-08Dated snapshotThinking and non-thinking
qwen3.7-max-previewqwen3.7-max-2026-05-17Thinking only
qwen3.7-plusqwen3.7-plus-2026-05-26Thinking and non-thinking
qwen3.7-flashqwen3.7-flash-2026-07-15Thinking and non-thinking

Calls go to Model Studio's OpenAI-compatible Chat Completions endpoint with a DashScope API key. The Qwen API guide covers the base URLs per region, rate limits, the enable_thinking parameter, and Claude Code setup.

Qwen 3.7 Benchmarks

Independent scores for Qwen 3.7 are thin, and most of them cover Max. Vals AI runs qwen3.7-max across its professional-task suite and reports a 58.80% average accuracy, with weights listed as private, a 984K context, and 66K max output tokens in its test setup.

Qwen 3.7 Max on Vals AI, checked November 18, 2026
BenchmarkScoreRank
LegalBenchHighest-ranked Vals result#28 of 151
Finance Agent (v2)47.78%53 of 81
Code Migration13.07%61 of 79

The Code Migration rank is the number to weigh for coding agents. For agentic coding, the November 2026 Terminal-Bench 4.1 board tracks newer Qwen builds such as Qwen3.8-Max-0902 rather than 3.7. Our coding model rankings compare the current options.

Where to Run Qwen 3.7 (Alibaba, OpenRouter, Fireworks)

Qwen 3.7 providers, checked November 18, 2026
ProviderTiersPrice per 1M (in / out)Notes
Alibaba Model StudioMax, Plus, Flash$2.50 / $7.50 (Max)First party, context-tier pricing, cache discounts
OpenRouterMax, Plus, Flash$1.475 / $4.425 (Max)Plus $0.32 / $1.28, Flash $0.03 / $0.13, 1M context listed
FireworksPlusSee Fireworks pricingServerless since its June 12, 2026 launch post

OpenRouter also lists a batch variant, qwen/qwen3.7-plus:batch, at $0.23/$0.92 with a 262K context. Fireworks says it serves Qwen 3.7 Plus on its own hardware under a partnership with Alibaba rather than forwarding requests.

Qwen 3.7 vs Qwen 3.8 and Qwen 4 Max

Qwen flagship and default tiers on Model Studio (International)
ModelPrice per 1M (in / out)Role in Alibaba's guide
qwen4-max$1.50 / $4.50Highest reasoning and understanding requirements
qwen3.8-max$2.00 / $6.00Previous flagship
qwen3.7-max$2.50 / $7.50Older flagship
qwen3.7-plus$0.40 / $1.60 up to 256KRecommended starting model
qwen3.8-flash$0.15 / $0.47Recommended for cost

On list price, qwen3.7-max is now the most expensive Max tier and qwen4-max is the cheapest. A new project that wants a Max model has little reason to pick 3.7. qwen3.7-plus is still the tier Alibaba points new users to.

Is Qwen 3.7 Open Source?

No. Alibaba has not published Qwen 3.7 weights, so it runs only on Alibaba's API and the providers that resell or host it. If you need weights for a private deployment, the open Qwen line is Qwen 3.8 27B and the Qwen 3.6 open models. Morph does not serve Qwen 3.7. Morph serves open-weight coding models such as Kimi K3, GLM-5.3, and DeepSeek V4.1 Flash on its own kernels; see the model list.

Qwen 3.7 FAQ

What is Qwen 3.7?

Qwen 3.7 is a proprietary model series from Alibaba in three tiers. qwen3.7-max is the text-only flagship, released May 20, 2026. qwen3.7-plus takes text, image, and video input (snapshot May 26, 2026). qwen3.7-flash is the vision-language tier (snapshot July 15, 2026). All three have a 1M-token context and are served through Alibaba Cloud Model Studio's API.

How much does Qwen 3.7 cost?

On Alibaba Cloud Model Studio (International), per million tokens: qwen3.7-max is $2.50 input and $7.50 output up to 1M tokens. qwen3.7-plus is $0.40/$1.60 up to 256K input tokens and $1.20/$4.80 from 256K to 1M, with a limited-time 20% discount. qwen3.7-flash is $0.03/$0.13 up to 32K, $0.10/$0.40 to 256K, and $0.20/$0.80 to 1M, and batch calls are half price.

When was Qwen 3.7 released?

Alibaba's snapshot IDs date the tiers. qwen3.7-max-preview maps to qwen3.7-max-2026-05-17, and the stable qwen3.7-max points to qwen3.7-max-2026-05-20, the date Vals AI lists as its release. qwen3.7-plus points to qwen3.7-plus-2026-05-26 and qwen3.7-flash to qwen3.7-flash-2026-07-15. A newer Max snapshot, qwen3.7-max-2026-06-08, is also listed.

Is Qwen 3.7 open source?

No. Qwen 3.7 is closed-weight: Vals AI lists its weights as private and Alibaba serves it only through its API. For open weights you can run yourself, use Qwen 3.8 27B or the Qwen 3.6 open models instead.

What is the difference between Qwen 3.7 Max and Qwen 3.7 Plus?

qwen3.7-max is the text-only flagship at a flat $2.50/$7.50 per million tokens. qwen3.7-plus accepts images and video as well as text, costs $0.40/$1.60 up to 256K input tokens, and supports up to 128K output tokens. Alibaba recommends qwen3.7-plus as the starting model and points harder reasoning work to qwen4-max.

Where can I use the Qwen 3.7 API?

First-party access is Alibaba Cloud Model Studio's OpenAI-compatible endpoint with the model IDs qwen3.7-max, qwen3.7-plus, and qwen3.7-flash. OpenRouter lists qwen/qwen3.7-max at $1.475/$4.425, qwen/qwen3.7-plus at $0.32/$1.28, and qwen/qwen3.7-flash at $0.03/$0.13 per million tokens. Fireworks serves Qwen 3.7 Plus on its serverless API.

Private deployments

The fastest endpoints are private deployments

Morph's top speeds come from dedicated deployments, not shared public endpoints: speculators trained on your traffic, caching tuned to your workload, and volume discounts over public per-token rates. Over 1 trillion tokens per day run this way.

Talk to us about a private deployment

Run open-weight coding models fast

Morph serves Kimi K3, GLM-5.3, and DeepSeek V4.1 Flash behind one OpenAI-compatible API.

Sources