What is GLM-5.2?
GLM-5.2 was Z.ai's flagship until GLM-5.3 arrived on the same base model. It was the first GLM to hold a 1M-token context, reads and writes text only, and added effort levels that trade depth for speed and cost; unlike GLM-5.3, it also lets you switch thinking off.
Z.ai still sells it on its API at the same price as GLM-5.3 and lists it in the chat app as the previous flagship, but the GLM Coding Plan now sends GLM-5.2 requests to GLM-5.3. Its main remaining draw is the MIT licence on its weights.
Is GLM-5.2 still available?
Yes on Z.ai's API, in its chat app and as open weights, but not on the GLM Coding Plan, which sends GLM-5.2 requests to GLM-5.3.
- API
- Listed as glm-5.2 at $1.4 / $4.4 per million tokens, next to GLM-5.3.
- Chat app
- In the chat.z.ai model picker as the previous flagship model.
- GLM Coding Plan
- Not served; requests for it go to GLM-5.3.
- Weights
- Still on Hugging Face and ModelScope under MIT; the model card points to GLM-5.3 as the newer version.
Who is GLM-5.2 best for?
GLM-5.2 is best for teams that need MIT-licensed weights, want to switch thinking off, or already run pipelines tuned for it.
Self-hosting under MIT
The weights are on Hugging Face and ModelScope under the MIT licence, with no regional limits and none of the review GLM-5.3's licence asks of the largest model hosts.
Fast answers with thinking off
It is the newest GLM flagship where you can disable thinking, which GLM-5.3 and GLM-5.3-Flash no longer allow.
Pipelines already tuned for it
Teams with prompts and tests built on GLM-5.2 can keep calling glm-5.2 on the API at the same price while they try GLM-5.3.
What's included with GLM-5.2?
A 1M-token context, effort levels with thinking you can turn off, efficient long-context attention, tool calling and MIT-licensed open weights.
- 1M-token context
- Z.ai trained it on long coding-agent sessions so it stays on track across the whole window, up from 200K in GLM-5.1.
- Effort levels
- Choose how hard it reasons on each request, trading quality against speed and token cost, or turn thinking off.
- Efficient long-context attention
- Z.ai's IndexShare design cuts compute per token by 2.9x at a 1M-token context.
- Tool use on the API
- Function calling, streaming, structured JSON output, context caching and MCP tools.
- Open weights
- MIT-licensed weights that run on Transformers, vLLM, SGLang, xLLM and KTransformers.
How much does GLM-5.2 cost?
GLM-5.2 costs $1.4 per million input tokens and $4.4 per million output tokens on Z.ai's API, the same as the newer GLM-5.3, and its MIT-licensed weights are free to download.
- Cached input $0.26 per million tokens
- 1M-token context, up to 128K tokens out
- Text input only
Prices are in US dollars. The GLM Coding Plan (from $18 a month) no longer serves GLM-5.2: it sends those requests to GLM-5.3, which costs the same per token on the API. You can still pick GLM-5.2 in Z.ai's free chat app at chat.z.ai. Full pricing on z.ai. We checked these prices on .
How do readers rate GLM-5.2?
No one has rated GLM-5.2 yet. Be the first.
Sign in to rate. One rating per account.
GLM-5.2 pros and cons
It codes close to Opus-class models for less, but it thinks at length, reads text only, and GLM-5.3 beats it at the same price.
Pros
- Users rate its coding close to Claude Opus 4.6 to 4.8 for everyday work, at a fraction of the per-token price.
- At high effort it uses about half the tokens of max with little loss in quality, several users report.
- Shows its full reasoning, which users like for spotting a run that is going wrong and steering it.
- MIT-licensed weights mean you are not tied to Z.ai: users run it through many third-party hosts, including European ones.
Cons
- Thinks at length at max effort: users report 15-minute reasoning runs, looping self-doubt, weekly plan quotas gone in two days and bills higher than they expected.
- Text only, so it cannot read screenshots or images; users pair it with a vision model for that.
- GLM-5.3 scores higher on every test in Z.ai's table at the same price, and the GLM Coding Plan no longer serves GLM-5.2.
- Some users say it ignores project rules they set and makes odd file edits in some coding tools.
Is GLM-5.2 open source?
Yes. GLM-5.2's code is public under the MIT licence, so you can run it on your own server.
What are the best GLM-5.2 alternatives?
GLM-5.2 is an open source alternative to MiniMax M3 and Grok 4.5. DeepSeek V4.1 Flash does the same job and is open source too.
MiniMax M3FreeMiniMax's multimodal coding and agent model with a 1M context and computer use.
Grok 4.5PaidxAI's coding and agent model with a 500K context and image input on the API.
DeepSeek V4.1 FlashOpen SourceChoose it when you want MIT weights, thinking you can turn off and image input at the lowest price: $0.30 / $1.20 per million tokens at peak, half that off-peak.
GLM-5.3FreeChoose it when you do not need MIT weights or thinking you can switch off; it runs on the same base, costs the same $1.4 / $4.4 and scores higher on every test in Z.ai's table.
GLM-5.3-FlashOpen SourceChoose it when you want MIT weights plus image and video input for $0.15 / $0.50 per million tokens; Z.ai rates it stronger than GLM-5.2.
Questions about GLM-5.2
What is GLM-5.2?
GLM-5.2 is Z.ai's previous flagship open-weights model, the first GLM with a 1M-token context, built for long coding agent runs and released under the MIT licence.
Who is GLM-5.2 for?
GLM-5.2 is best for teams that need MIT-licensed weights, want to switch thinking off, or already run pipelines tuned for it.
Is GLM-5.2 free?
Partly. GLM-5.2's weights are free on Hugging Face under the MIT license, and you can pick it in Z.ai's free chat app at chat.z.ai. Through Z.ai's API it costs $1.4 per million input tokens and $4.4 per million output tokens.
How much does GLM-5.2 cost?
GLM-5.2's paid plans are API at $1.4 / $4.4 per million input / output tokens.
Is GLM-5.2 open source?
Yes. GLM-5.2's code is public under the MIT licence, so you can run it on your own server.
What are the best GLM-5.2 alternatives?
The closest alternatives on Recaply are MiniMax M3, Grok 4.5, DeepSeek V4.1 Flash, GLM-5.3 and GLM-5.3-Flash.
Where can I try GLM-5.2?
On its official website, z.ai.
Should I use GLM-5.2 or GLM-5.3?
For most coding work, GLM-5.3. Z.ai built it on the same base model, reports 50% better results on its own coding benchmark, and charges the same $1.4 / $4.4 per million tokens. Stay on GLM-5.2 only if you need MIT-licensed weights or want to turn thinking off.
Can I use GLM-5.2 on the GLM Coding Plan?
No. The GLM Coding Plan sends requests for GLM-5.2 or GLM-5.1 to GLM-5.3 automatically. To keep using GLM-5.2 itself, call glm-5.2 on Z.ai's pay-as-you-go API, pick it in the chat app, or host the weights yourself.
Can I turn off thinking in GLM-5.2?
Yes. Set the thinking type to disabled in the API request; thinking is on by default. GLM-5.3 and GLM-5.3-Flash always think, and Z.ai tells anyone moving from GLM-5.2 to switch to low effort instead.
Our verdict
Most people should use GLM-5.3 instead; keep GLM-5.2 only for a specific reason.
GLM-5.3 runs on the same base, costs the same per token and scores higher on every test in Z.ai's table, while GLM-5.2 thinks at length and reads text only.
- Skip it for
- New projects that need neither MIT-licensed weights nor thinking you can switch off.
- Check first
- If you self-host for the MIT licence, check whether GLM-5.3-Flash, which is also MIT licensed, already does the job.
Chroma Context-1
Kimi K2.7 Code
MiMo-V2.5-Pro
GLM-4.7
North Mini Code
Laguna S 2.1