Which Inference Providers Host GLM 5.2? 2026 List
Nov 19, 2026

Which Inference Providers Host GLM 5.2? 2026 List

Which inference providers host GLM 5.2? 23+ hosts list it, but context limits, quantization and prices differ. See who serves it and how to choose safely.

Which inference providers host GLM 5.2: context, quantization and price differences

Two providers both advertise "GLM 5.2." One gives you the full 1M-token context at 8-bit precision. The other caps you at about 200K tokens and runs a 4-bit build. The model name on the invoice is the same. The model you're actually getting is not.

Which inference providers host GLM 5.2? Plenty. On November 19, 2026, OpenRouter's public endpoint list showed 33 GLM 5.2 endpoints from 23 providers. Fireworks, Z.ai's own API and GLM5 also serve it directly. The useful question is which host serves it the way your workload needs, and that comes down to four things: context limit, quantization, output cap and price.

This guide lists the hosts we could verify, explains what differs between them, and gives a checklist for choosing. Provider data comes from OpenRouter's public endpoints API and the providers' own model pages, retrieved November 19, 2026. Model facts come from Z.ai's official documentation. We did not run latency or quality tests on each host. GLM5 publishes this guide and is itself one of the hosts mentioned, so weigh our own listing accordingly.

GLM 5.2 Hosts at a Glance

Host typeExamples (verified Nov 19, 2026)Good for
First-partyZ.ai APIReference behaviour, official pricing, full 1M context
Serverless inference cloudsTogether, Fireworks, DeepInfra, Novita, SiliconFlow, Nebius, Baseten, Parasail, FriendliPay-per-token, quick start, OpenAI-compatible APIs
GPU clouds and platformsCoreWeave, DigitalOcean, Cloudflare, GMICloud, AtlasCloudTeams already on that cloud
Large platformsAlibaba, Baidu, MistralEnterprise contracts and regional presence
Specialists and aggregatorsWafer, Relace, Morph, Ambient, StreamLake, Venice; OpenRouter as a routerPrice shopping, niche features, failover
App-level APIsGLM5 (glm-5.2 at https://glm5.app/api/v1)One key for several GLM and non-GLM models

The full OpenRouter provider list for GLM 5.2 on that date was: Alibaba, Ambient, AtlasCloud, Baidu, BaseTen, Cloudflare, CoreWeave, DeepInfra, DigitalOcean, Friendli, GMICloud, Mistral, Morph, Nebius, Novita, Parasail, Relace, SiliconFlow, StreamLake, Together, Venice, Wafer and Z.AI. Separately, Fireworks publishes its own GLM 5.2 model page with the ID accounts/fireworks/models/glm-5p2.

What Actually Differs Between GLM 5.2 Hosts

GLM 5.2 itself is defined by Z.ai's documentation: text in, text out, a 1M-token context and up to 128K output tokens. Hosts don't always serve all of that. In the OpenRouter endpoint data, these were the four places they diverged:

1. Context limit

Most endpoints advertised the full 1M (1,048,576) tokens. Some didn't: one endpoint was capped at 202,752 tokens and two others at 262,144. If your workload relies on GLM 5.2's headline feature, feeding it a large codebase or document set in one go, a capped host will either reject the request or force you to cut the input.

2. Quantization

Endpoints reported FP8 or FP4 precision, and several didn't disclose it. Z.ai's own endpoint reported FP8. Lower precision cuts serving cost and usually price, but it can change output quality, especially on long reasoning chains and exact code. Some of the cheapest endpoints were FP4.

Technical note: FP4 stores each weight in 4 bits, versus 8 bits for FP8. That halves the memory needed, which is why a 753B-parameter model becomes affordable to host. Modern FP4 formats can hold up well, but "well on average" isn't the same as "identical on your hardest prompt." Test before you switch.

3. Output cap

Maximum output varied from about 128K tokens up to over 900K on some endpoints. Most workloads never come close. If you generate long files or reports in one response, though, check this number.

4. Price

Z.ai's direct price is $1.40 per million input tokens and $4.40 per million output tokens. Third-party prices on OpenRouter ranged both below and above that. Some providers undercut it substantially, often with FP4 builds. Others matched it exactly. Fireworks' model page listed serverless pricing of $1.40 input, $0.14 cached input and $4.40 output per million tokens. Prices move often, so treat any number as a snapshot.

How to Choose a GLM 5.2 Provider

Use this checklist in order. Each step removes hosts that can't do the job:

  1. Context: does the endpoint support the longest prompt you'll realistically send? If you need more than 262K tokens, filter to full-1M endpoints.
  2. Precision: is FP4 acceptable for your task? For production coding agents, start with FP8 or the first-party endpoint and only move to FP4 after a side-by-side test.
  3. Features: confirm tool calling, structured output and streaming work the way your code expects. Not every host exposes every parameter the same way.
  4. Price: only now compare token prices among the hosts that passed steps 1–3.
  5. Reliability: pick a primary and a fallback. Routers like OpenRouter can fail over automatically, but your code should still log which provider served each request.

Rule of thumb: the cheapest GLM 5.2 endpoint is only cheap if it passes your evaluation set. Run the same 20 real prompts through your top two candidates before committing.

For a deeper look at running GLM reliably at scale, see the best way to run GLM in production.

Which Providers Host GLM 5.1?

GLM 5.1 has a smaller host list. OpenRouter showed 11 endpoints on November 19, 2026, from Chutes, StreamLake, SiliconFlow, AtlasCloud, Alibaba, Novita, Baidu, GMICloud, Friendli, Z.AI and Venice. All were close to GLM 5.1's 200K-token context and mostly FP8. Fireworks also lists it as accounts/fireworks/models/glm-5p1.

Can I run GLM 5.1 on Fireworks AI? Yes. Fireworks publishes a GLM 5.1 model page with that model ID. Check the page for whether serverless or on-demand deployment fits your account and volume.

Does Fireworks AI support GLM 4.5? Fireworks also lists a GLM-4.5 model page (accounts/fireworks/models/glm-4p5). For new projects, consider a newer GLM release. See what GLM is for how the versions compare.

A Shortcut: One API for GLM 5.2 and Its Siblings

If you'd rather not manage several provider accounts, GLM5's API serves glm-5.2 alongside glm-5.3, glm-5.1, glm-5 and glm-4.7 through one OpenAI-compatible endpoint. Switching models is a one-word change, which makes side-by-side testing easy. See model IDs and credit rates in the GLM5 API docs, or try GLM 5.2 in the browser first.

One thing to avoid: if you sign up for Z.ai's GLM Coding Plan to get GLM 5.2, know that Z.ai documents automatic routing of GLM-5.2 requests to GLM-5.3 on that plan. That's fine for daily coding, but it won't give you GLM 5.2 itself.

Frequently Asked Questions

Which inference providers host GLM 5.2?

As of November 19, 2026: Z.ai, Together, Fireworks, DeepInfra, Novita, SiliconFlow, Nebius, Baseten, Parasail, Friendli, CoreWeave, DigitalOcean, Cloudflare, GMICloud, AtlasCloud, Alibaba, Baidu, Mistral, Wafer, Relace, Morph, Ambient, StreamLake and Venice. OpenRouter routes across many of them, and GLM5 also serves it.

Do all GLM 5.2 providers support the 1M-token context?

No. Most endpoints we checked did, but some were capped at around 200K–262K tokens. Always check the endpoint's listed context length.

Why is GLM 5.2 cheaper on some providers?

Lower prices often come with FP4 quantization, smaller context caps or promotional pricing. Compare like for like.

What is the official GLM 5.2 API price?

Z.ai lists $1.40 per million input tokens, $0.26 per million cached input tokens and $4.40 per million output tokens.

Which providers host GLM 5.1?

Z.ai, Chutes, StreamLake, SiliconFlow, AtlasCloud, Alibaba, Novita, Baidu, GMICloud, Friendli and Venice on OpenRouter, plus Fireworks.

The Bottom Line

GLM 5.2 is widely hosted, so availability isn't the problem. Consistency is. Context caps, FP4 versus FP8 builds and price differences mean two "GLM 5.2" endpoints can behave quite differently. Filter by context and precision first, compare price last, and confirm with your own prompts.

The quickest way to start comparing is a single endpoint that serves several GLM versions: get a GLM5 API key and test glm-5.2 against glm-5.3 on your real workload.

Sources

Snapshot taken November 19, 2026. Provider lists, prices, context limits and quantization change frequently. Verify on each provider's page before you commit.

Start Using GLM 5 Today

Try GLM 5 free — reasoning, coding, agents, and image generation in one platform.