- Type
- Open weightsMIT License
- Params
- 753B
- Context
- 1M
active per word not recorded by us · about 786K words of context
Our take
Written Aug 6, 2026Available from 30 hosted offers, this large downloadable model from Zhipu carries a permissive MIT licence and a one-million-token request limit. It is a strong choice for teams that want top-tier capability without licensing restrictions.
Who should pick it
Pick this when you need top-tier capability with a genuinely permissive licence, or for long-context workloads on a budget. Use it if you want leverage over providers: 30 offers means real price competition. Skip it if you need image, audio or video input, or if you want to self-host on a single consumer GPU.
The case for it
- Permissive MIT licence allows commercial use, fine-tuning and redistribution.
- 30 current hosted offers, so price competition is unusually strong.
- One-million-token request limit matches the proprietary top-tier models.
The case against it
- Text-only; no image, audio or video input, unlike Gemini or Claude lines.
- Over 753.3 billion parameters puts even compressed weights beyond workstation reach.
00
How good is it?
An open-weights text model for everyday questions, drafting, coding and tool use.
Good at
- getting answers to everyday questionsArena Text (overall) · 23rd of 168
- drafts, rewrites and editingArena Creative Writing · 21st of 168
- writing and completing codeArena Coding · 37th of 168
- calling tools to carry out requestsArena Agent · Tool use · 9th of 55
- changing course when you give new instructionsArena Agent · Steerability · 12th of 55
EverydayGeneral questions and everyday reasoning
4 of 5
Arena Text (overall)23rd of 168 · 1476
Arena Hard Prompts 27th of 168Arena Maths 20th of 163LiveBench Mathematics 30th of 58LiveBench Data Analysis 34th of 58LiveBench Reasoning 45th of 58
CodingWriting and fixing code on its own
4 of 5
Arena Coding37th of 168 · 1512
Arena Code (WebDev) 18th of 95LiveBench Coding 19th of 58
AgenticPlanning, calling tools, staying on task
3 of 5
Arena Agent14th of 55 · 0.04
LiveBench Agentic Coding 32nd of 58
WritingDrafting and rewriting prose
3.5 of 5
Arena Creative Writing21st of 168 · 1455
LiveBench Language 38th of 58
How it behaves in an agent loop
Tool usereaches for the right one, and does not invent one9th of 55
Steerabilitydoes what it was asked, and changes course when told12th of 55
Recoverygets back on track after a command fails28th of 55
Task outcomefinishes what the session set out to do16th of 55
Placings on Arena's agent boards, from live sessions people ran themselves. A model can lead on one of these and sit mid-field on the others.
Other boards it appears on
Arena Instruction Following 25th of 168LiveBench 38th of 58LiveBench Instruction Following 47th of 58Arena Agent · Tool use 9th of 55Arena Agent · Steerability 12th of 55Arena Agent · Task outcome 16th of 55Arena Agent · Recovery 28th of 55
Boards this model appears on that none of the ratings above are built on.
Every published score for this model20 scoresEvery figure we hold, from 20 boards, with who ran it and a link to the source — including the boards no rating above is built on.
01
Can you run it yourself?
A card many people ownToo large
GeForce RTX 4090 · 24 GB
Weights at 475 / 24 GBest
Usable contextNot calculated for spilled setups
Decode speedNot estimated for spilled setups
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
One step upToo large
Apple M1 Pro (16-core GPU) · 32 GB
Weights at 475 / 32 GBest
Usable contextNot calculated for spilled setups
Decode speedNot estimated for spilled setups
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
One step downToo large
Radeon RX 7900 XT · 20 GB
Weights at 475 / 20 GBest
Usable contextNot calculated for spilled setups
Decode speedNot estimated for spilled setups
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Your hardware
Checking your profile…
Memory use by level
Against a 24 GB card.
What is quantisation? →475 GBest
557.2 GBest
832.4 GBest
This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
Check against your own machine → · Where to rent it hosted →
02
Or rent it from someone else
Prices checked between 4 hours and 3 days ago — each listing carries its own date.
Some hosts sell this model at two prices: on their own price list (“direct”) and on their OpenRouter listing (“through OpenRouter”). Where the two differ, the row shows both, each with the date we last read it.
- per 1M tokens
- $0.46 in / $0.91 out
- Context served
- 1M
- Throughput
- ~54 tok/s
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| RekaThrough OpenRouter | $0.46 / $0.91checked 4 hours ago | 1M131K max reply | 54 tok/s | No | No | Confirmed |
| StreamLakefp8Through OpenRouter | $0.56 / $1.76checked 4 hours ago | 1M128K max reply | 52 tok/s | No | Yesunknown period | Unknown |
| DeepInfrafp4Direct and through OpenRouter | $0.75 / $2.40directchecked 4 hours ago$0.56 / $1.80through OpenRouterchecked 4 hours ago | 1M164K max reply through OpenRouter | 53 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| Novita AIfp8Direct and through OpenRouter | $1.40 / $4.40directchecked 4 hours ago$0.65 / $2.04through OpenRouterchecked 4 hours ago | 1M131K max reply through OpenRouter | 70 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| SiliconFlowfp8Through OpenRouter | $0.70 / $2.20checked 4 hours ago | 1M262K max reply | 71 tok/s | No | No | Confirmed |
| DigitalOcean GradientThrough OpenRouter | $0.70 / $2.20checked 4 hours ago | 262K236K max reply | 85 tok/s | No | No | Confirmed |
| Decartmxfp4Through OpenRouter | $0.39 / $2.40checked 4 hours ago | 1M944K max reply | 59 tok/s | No | No | Confirmed |
| CoreWeavefp4Through OpenRouter | $0.76 / $2.42checked 4 hours ago | 1M944K max reply | 8 tok/s | No | No | Confirmed |
| Morphfp8Through OpenRouter | $0.46 / $2.84checked 4 hours ago | 1M944K max reply | 76 tok/s | No | No | Confirmed |
| AtlasCloudfp8Through OpenRouter | $0.94 / $2.95checked 4 hours ago | 1M131K max reply | 61 tok/s | No | Yesunknown period | Unknown |
| Phalafp8Through OpenRouter | $1.26 / $3.00checked 4 hours ago | 1M131K max reply | 122 tok/s | No | No | Confirmed |
| Alibaba Cloudfp8Through OpenRouter | $0.97 / $3.04checked 4 hours ago | 1M131K max reply | 58 tok/s | No | Yesunknown period | Unknown |
| OpenRouterOpenRouter's own listing | $0.41 / $3.99checked 4 hours ago | 1M | not measured | Unknown | Unknown | Unknown |
| WaferThrough OpenRouter | $0.41 / $3.99checked 4 hours ago | 1M944K max reply | 85 tok/s | No | No | Confirmed |
| RelaceThrough OpenRouter | $0.13 / $4.00checked 4 hours ago | 1M131K max reply | 80 tok/s | No | No | Confirmed |
| Inceptronfp4Through OpenRouter | $1.39 / $4.39checked 4 hours ago | 1M944K max reply | 59 tok/s | No | No | Confirmed |
| Together AIThrough OpenRouter | $1.40 / $4.40checked 4 hours ago | 1M944K max reply | 67 tok/s | No | No | Confirmed |
| Parasailfp4Through OpenRouter | $1.40 / $4.40checked 4 hours ago | 262K236K max reply | 60 tok/s | No | No | Confirmed |
| Mistral AInvfp4Through OpenRouter | $1.40 / $4.40checked 4 hours ago | 1M131K max reply | 93 tok/s | No | Yes30 days | Unknown |
| Cloudflare Workers AIThrough OpenRouter | $1.18 / $4.40checked 4 hours ago | 262K236K max reply | 23 tok/s | No | Yesunknown period | Unknown |
| GMICloudfp8Through OpenRouter | $1.40 / $4.40checked 4 hours ago | 1M944K max reply | 15 tok/s | No | Yesunknown period | Unknown |
| Baidufp8Through OpenRouter | $1.40 / $4.40checked 22 hours ago | 1M131K max reply | 57 tok/s | No | Yesunknown period | Unknown |
| Fireworks AIThrough OpenRouter | $1.40 / $4.40checked 3 days ago | 1M944K max reply | 75 tok/s | No | No | Confirmed |
| FriendliThrough OpenRouter | $1.40 / $4.40checked 4 hours ago | 1M944K max reply | 83 tok/s | No | Yesunknown period | Unknown |
| Z.AIfp8Through OpenRouter | $1.40 / $4.40checked 4 hours ago | 1M131K max reply | 68 tok/s | No | No | Confirmed |
| Basetenfp8Through OpenRouter | $1.40 / $4.40checked 4 hours ago | 1M262K max reply | 49 tok/s | No | No | Confirmed |
| Venice AIfp8Through OpenRouter | $1.40 / $4.40checked 4 hours ago | 1M131K max reply | 91 tok/s | No | No | Confirmed |
| Mistral AIeuThrough OpenRouter | $1.54 / $4.84checked 4 hours ago | 1M128K max reply | 89 tok/s | No | Yes30 days | Confirmed |
| Baidufast tierfp4Through OpenRouter | $1.40 / $4.90checked 34 hours ago | 1M131K max reply | 81 tok/s | No | Yesunknown period | Unknown |
| Basetenfast tierfp8Through OpenRouter | $2.10 / $6.60checked 4 hours ago | 1M262K max reply | 109 tok/s | No | No | Unknown |
| Alibaba Cloudfast tierfp8Through OpenRouter | $2.31 / $7.26checked 4 hours ago | 1M131K max reply | 48 tok/s | No | Yesunknown period | Unknown |
| Decartfast tierfp4Through OpenRouter | $2.25 / $8.00checked 4 hours ago | 1M944K max reply | 162 tok/s | No | No | Unknown |
DeepInfra
Novita AI
DigitalOcean Gradient
Morph
OpenRouter
Together AI
Mistral AI
Cloudflare Workers AI
Baidu
Fireworks AIAcross the 32 listings we hold: 31 say they do not train on prompts (2 of them only through OpenRouter), 0 say they do and 1 does not say. 19 appear in the zero-retention registry we check (2 of them only through OpenRouter); the rest are unknown to us.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| RekaThrough OpenRouter | ✓ | ✓ | ✓ |
| StreamLakefp8Through OpenRouter | ✓ | ✓ | ✓ |
| DeepInfrafp4Direct and through OpenRouter | ✓ | ✓ | ✓ |
| Novita AIfp8Direct and through OpenRouter | ✓ | ✓ | ✗ |
| SiliconFlowfp8Through OpenRouter | ✓ | ✗ | ✗ |
| DigitalOcean GradientThrough OpenRouter | ✓ | ✓ | ✓ |
| Decartmxfp4Through OpenRouter | ✓ | ✓ | ✓ |
| CoreWeavefp4Through OpenRouter | ✓ | ✓ | ✓ |
| Morphfp8Through OpenRouter | ✓ | ✓ | ✓ |
| AtlasCloudfp8Through OpenRouter | ✓ | ✓ | ✗ |
| Phalafp8Through OpenRouter | ✓ | ✓ | ✓ |
| Alibaba Cloudfp8Through OpenRouter | ✓ | ✓ | ✓ |
| OpenRouterOpenRouter's own listing | ✓ | ✓ | ✓ |
| WaferThrough OpenRouter | ✓ | ✓ | ✓ |
| RelaceThrough OpenRouter | ✓ | ✓ | ✗ |
| Inceptronfp4Through OpenRouter | ✓ | ✓ | ✓ |
| Together AIThrough OpenRouter | ✓ | ✓ | ✓ |
| Parasailfp4Through OpenRouter | ✓ | ✓ | ✓ |
| Mistral AInvfp4Through OpenRouter | ✓ | ✓ | ✓ |
| Cloudflare Workers AIThrough OpenRouter | ✓ | ✓ | ✓ |
| GMICloudfp8Through OpenRouter | ✓ | ✓ | ✓ |
| Baidufp8Through OpenRouter | ✓ | ✓ | ✓ |
| Fireworks AIThrough OpenRouter | ✓ | ✓ | ✓ |
| FriendliThrough OpenRouter | ✓ | ✓ | ✓ |
| Z.AIfp8Through OpenRouter | ✓ | ✓ | ✗ |
| Basetenfp8Through OpenRouter | ✓ | ✓ | ✓ |
| Venice AIfp8Through OpenRouter | ✓ | ✓ | ✓ |
| Mistral AIeuThrough OpenRouter | ✓ | ✓ | ✓ |
| Baidufast · fp4Through OpenRouter | ✓ | ✓ | ✓ |
| Basetenfast · fp8Through OpenRouter | ✓ | ✓ | ✓ |
| Alibaba Cloudfast · fp8Through OpenRouter | ✓ | ✓ | ✓ |
| Decartfast · fp4Through OpenRouter | ✓ | ✓ | ✓ |
Tool calling: 32 of 32 listings say yes. JSON output: 31 of 32 listings say yes, 1 says no. Strict schema: 27 of 32 listings say yes, 5 say no.
03
Models people weigh against GLM 5.2
SucceedsGLM 5.1
04
When we formed this view
Recent changes
Oct 1, 2026Price changeGLM 5.2 repriced across 5 hosts: Wafer input down 71%, Inceptron input up 85%
What moved
GLM 5.2 moved on 5 hosts: Inceptron: input +85% ($0.75 → $1.39 per 1M tokens), output +83% ($2.40 → $4.39 per 1M tokens), cache read +48% ($0.169 → $0.250 per 1M tokens); Wafer: input −71% ($1.40 → $0.41 per 1M tokens), output −9% ($4.40 → $3.99 per 1M tokens); Reka: input −50% ($1.40 → $0.70 per 1M tokens), output −68% ($4.40 → $1.40 per 1M tokens), cache read −30% ($0.20 → $0.14 per 1M tokens); Morph: input +20% ($0.387 → $0.464 per 1M tokens), output +6% ($2.69 → $2.84 per 1M tokens); Relace: input −3% ($0.135 → $0.131 per 1M tokens), cache read −3% ($0.135 → $0.131 per 1M tokens)Sep 30, 2026Price changeGLM 5.2 cut across 3 hosts, by up to 72% at Decart (mxfp4) (input)
What moved
GLM 5.2 moved on 3 hosts: Decart (mxfp4): input −72% ($1.40 → $0.39 per 1M tokens), output −45% ($4.40 → $2.40 per 1M tokens), cache read −42% ($0.26 → $0.15 per 1M tokens); Relace: input −33% ($0.200 → $0.135 per 1M tokens), cache read −33% ($0.200 → $0.135 per 1M tokens); Reka: cache read −23% ($0.26 → $0.20 per 1M tokens)Sep 29, 2026Price changeGLM 5.2 repriced across 2 hosts: Relace input down 43%, Inceptron input up 110%
What moved
GLM 5.2 moved on 2 hosts: Inceptron: input +110% ($0.3565 → $0.7500 per 1M tokens), output −0.0% ($2.401 → $2.400 per 1M tokens), cache read +1% ($0.167 → $0.169 per 1M tokens); Relace: input −43% ($0.35 → $0.20 per 1M tokens), output +14% ($3.50 → $4.00 per 1M tokens)Sep 28, 2026Price changeGLM 5.2 repriced across 4 hosts: Inceptron input down 27%, Wafer input up 30%
What moved
GLM 5.2 moved on 4 hosts: Wafer: input +30% ($1.40 → $1.82 per 1M tokens); Inceptron: input −27% ($0.49 → $0.36 per 1M tokens), output +25% ($1.925 → $2.401 per 1M tokens), cache read +22% ($0.137 → $0.167 per 1M tokens); Cloudflare: input −16% ($1.40 → $1.18 per 1M tokens); Relace: input −13% ($0.40 → $0.35 per 1M tokens)Sep 26, 2026Price changeHost Inceptron cut GLM 5.2 output pricing by 40%
What moved
input −24% ($0.646 → $0.489 per 1M tokens), output −40% ($3.18 → $1.92 per 1M tokens), cache read −35% ($0.208 → $0.136 per 1M tokens)Sep 25, 2026Price changeHost Inceptron repriced GLM 5.2: input down 28%, cache read up 11%
What moved
input −28% ($0.90 → $0.65 per 1M tokens), cache read +11% ($0.19 → $0.21 per 1M tokens)Each date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- We do not hold the active parameter count for it, so how much of it runs on any one token is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 1 of 32 listings does not say whether it trains on prompts, and 2 answer only through OpenRouter, not for their own listing.
- We hold no batch or off-peak rate for any of its listings.
- We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
05
Licence and identifiers
What the licence allowsMIT License, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
MIT License
Open, few conditionsCommercial use allowed
Fully permissive: do anything with attribution. No patent grant, unlike Apache-2.0.
Identifiers
- Hugging Face
- zai-org/GLM-5.2
- Architecture
- Mixture of experts
- Takes in, gives back
- Text in, text out
- Catalogue slug
- z-ai-glm-5-2