Free Qwen, GLM and Kimi API in 2026: Where They Are Free

Qwen, GLM and Kimi are three Chinese open-weight model families, and each vendor treats "free" differently. Alibaba (Qwen) gives new Model Studio accounts a token quota. Z.AI (GLM) prices a few Flash models at $0. Moonshot (Kimi) has no free tier at all. Everywhere else you reach these models through a host that gives a daily limit, a monthly allowance or a one-time credit. Every vendor statement below was read on the vendor's own page on 2026-11-05, and every limit for a host comes from the catalog row that links to it.

Qwen: Alibaba Cloud Model Studio #

Alibaba's free quota page says each model has its own free quota, "typically 1,000,000 tokens", and the quota cannot be shared or moved between models. Input and output tokens count against the same total. What to know before you rely on it:

• 90 days. The clock starts at the latest of activation, model release or approval of your model request. Unused quota is voided at expiry and not reissued.
• Singapore only. Only models in the Singapore region with International deployment scope qualify, and you must complete your account information before activating Model Studio.
• The list of models changed. The page has a "Removed free quotas" list that includes qwen-plus, qwen-max, qwen3.6-plus, qwen3.5-plus, qwen-flash and qwen-turbo, so new users no longer receive quota for them; existing quota stays valid until it expires. New models still get quota when they launch, and the catalog row lists qwen3.8-max, which is not on the removed list (page read 2026-11-05). Open the model's details page in the console to confirm it still carries a quota before you plan around it.
• Overage is billed by default. There is a "Free Quota Only" switch, off by default. With it on, an exhausted model returns the error `AllocationQuota.FreeTierOnly` and you are not charged. With it off, extra usage is billed pay-as-you-go. Turn it on first if you want a hard stop.
• Real-time inference only. Batch calls, fine-tuning and deployment are not covered, and quota does not move to another model on its own.

Alibaba Cloud also runs a broader free trial with credits on 80+ products.

GLM: Z.AI and BigModel #

Z.AI's pricing page prices three models at Free for input, output and cached input: GLM-4.7-Flash and GLM-4.5-Flash (text) and GLM-4.6V-Flash (vision). The page publishes no free-tier rate limit or concurrency number; the catalog row notes that concurrency is set by the platform per plan. The newer GLM 5.x models are not on the free list, so for those you need a host that gives credits (below).

Zhipu's China platform, BigModel, marks more Flash models as free in the catalog: GLM-4.7-Flash (200K context, 128K max output), GLM-4.5-Flash, GLM-4-Flash-250414, the vision models GLM-4.6V-Flash, GLM-4.1V-Thinking-Flash and GLM-4V-Flash, plus image and video Flash models. Its limits are unpublished and the site is aimed at China, so treat it as a second option, not the default.

Kimi: no free tier #

Moonshot's rate limit page mentions no free tier. It says "a minimum recharge of $1 is required to start using the API", and that once your cumulative recharge reaches $5 you get a $5 voucher (vouchers do not count toward the cumulative total). The rate limit table did not load when we read the page, so we do not quote request or concurrency numbers here. In short: Kimi's own API is pay as you go from a $1 top-up. Free access to Kimi K3 or K2 comes through hosts that include them in a credit: EmberCloud for Startups ($100 for startups and builders, lists `kimi-k3`), SmarToken ($5 trial credit, lists `kimi-k2`), and the startup programs of Together AI, Fireworks, Baseten and DeepInfra. Those programs have eligibility rules and are not open to everyone.

Where these models are free or free to start #

CleanAPIs also lists Qwen3.8-27B at 0.5M free tokens a month and 80 requests a minute, but its catalog row is flagged as changed, so check its page first.

Which one to pick #

• Try Qwen with the biggest budget: IteraCompute (30M tokens a month) or Alibaba's own 1M-token quota per model, if the model you want still has one.
• Highest request rate, small outputs: OVHcloud (400 requests a minute) or Cloudflare (300 a minute inside the daily Neuron budget).
• GLM at $0 on the vendor itself: Z.AI's GLM-4.7-Flash or GLM-4.5-Flash. Pick GLM-4.6V-Flash if you need image input.
• Kimi K3: there is no free path for everyone. A small credit via SmarToken or EmberCloud is the nearest, or the $1 top-up on Moonshot's platform.
• Long context: GLM-4.7-Flash lists 200K; Qwen3.8-27B on IteraCompute is 128K and 256K on OVHcloud and Scaleway.

Setup in four steps #

1. Pick a row above and create the account. Most need only an email or Google sign-in.
2. Create an API key in the provider's console.
3. Point any OpenAI-compatible SDK at the provider's base URL and use the model ID from its page. IDs differ by provider (Groq: `qwen/qwen3.8-27b`; Cerebras: `qwen-3.8-27b`).
4. Set a spending guard before you call anything: Alibaba's "Free Quota Only" switch, or a payment method left off wherever it is optional.

The code is the same everywhere; only the base URL, key and model ID change, so you can move between hosts as quotas run out.

Data residency and terms #

The vendor pages we read do not state where free-tier requests are processed or how prompts are kept, so we do not claim anything here. Alibaba's free quota is tied to the Singapore region; Z.AI and BigModel are separate platforms (international and China); AMD Radeon Cloud and BigModel are China-site services. Read each provider's terms before sending customer data, and do not send private data to a free tier you have not read the terms of.

Running them yourself #

Qwen and GLM have open-weight releases, and Mistral's own model page lists GLM 5.3 and GLM 5.3 Flash as MIT-licensed third-party models. If you have the hardware, the guide to running AI locally for free covers the options; for hosted open-weight DeepSeek see the free DeepSeek API guide.

Limits and the catch #

• Quotas expire. Alibaba's lapses after 90 days, Cerebras's credit after 30.
• Credits are not free models. AkashML, EmberCloud and SmarToken give a balance you spend, not an unlimited model.
• Free lineups rotate. Alibaba already removed models from its free list; keep the model ID in config.
• Related guides: free reasoning model API, free OpenRouter models, free LLM APIs compared.