- Size
- 753B parameterscounted from the safetensors header on the Hub
- Licence
- unknown, MIT LicenseOur reading, check the licence: You can use, change and redistribute the model for any purpose, including commercially. Keep the copyright notice and the licence text with copies. You do not have to share your changes.
- On the Hub since
- 16 Jun 2026the date the repo was created
- Context length
- 1M tokensfrom the model's config
- Kind
- Base model (no base model declared on the Hub)GlmMoeDsaForCausalLM
- On Hugging Face
- zai-org/GLM-5.2311k downloads in the last 30 days, 5.2k likes
Scores
- effort max22.0%±4.214 Oct 2026, log
- effort none19.0%±3.914 Oct 2026, log
- effort low18.0%±3.914 Oct 2026, log
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5-2_max, 79.0% (Anthropic). All 166 results
- effort max9.5%29 Jun 2026, Epoch AI
Epoch's name: glm-5.2. Best closed model on this set: gpt-6.2-astra_max, 92.4% (OpenAI). All 33 results
- effort max29.3%±7.219 Jun 2026, log
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5-2_max, 100.0% (Anthropic). All 83 results
- effort max59.2%±3.019 Jun 2026, Epoch AI
- effort low54.7%±3.029 Aug 2026, log
- effort none42.5%±2.929 Aug 2026, log
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5-2_max, 95.1% (Anthropic). All 127 results
- effort max91.9%±1.624 Jun 2026, log
- effort low87.9%±2.310 Aug 2026, log
- effort none71.2%±3.210 Aug 2026, log
Epoch's name: glm-5.2. Best closed model on this set: gpt-6.2-astra_max, 96.8% (OpenAI). All 341 results
- effort max36.7%±11.521 Oct 2026, Epoch AI
Epoch's name: glm-5.2. Best closed model on this set: claude-sonnet-5-5_max, 97.8% (Anthropic). All 47 results
- effort low19.0%±3.927 Aug 2026, log
- effort minimal18.0%±3.927 Aug 2026, log
- effort none18.0%±3.927 Aug 2026, log
- effort medium15.0%±3.627 Aug 2026, log
Epoch's name: glm-5.2. Best closed model on this set: gpt-6.2-astra_max, 95.0% (OpenAI). All 148 results
- effort max86.4%±4.225 Jun 2026, log
- effort low75.6%±6.510 Aug 2026, log
- effort none28.9%±6.810 Aug 2026, log
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5_high, 100.0% (Anthropic). All 318 results
- effort max34.2%±1.527 Aug 2026, log
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5-2_max, 94.0% (Anthropic). All 105 results
- effort max78.7%±1.925 Jun 2026, log
Epoch's name: glm-5.2. Best closed model on this set: claude-opus-4-7_max, 83.5% (Anthropic). All 35 results
- effort unknown45.2%via Epoch AI (error bar left out)
Epoch's name: glm-5.2. Best closed model on this set: gemini-4-carbon_unknown, 84.2% (Google DeepMind). All 66 results
- effort unknown77.0%arcprize.org
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5_max, 98.5% (Anthropic). All 318 results
- effort unknown22.8%via Epoch AI
Epoch's name: glm-5.2. Best closed model on this set: gpt-6-astra_max, 95.0% (OpenAI). All 299 results
- effort max20.9%via Epoch AI
- effort none3.1%via Epoch AI
Epoch's name: glm-5.2. Best closed model on this set: gemini-4-carbon_high, 36.0% (Google DeepMind). All 271 results
- effort max; Harness: mini-swe-agent; Reasoning effort: max43.8%42.1 to 45.5deepswe.datacurve.ai
- effort high; Harness: mini-swe-agent; Reasoning effort: high36.3%31.5 to 41.0deepswe.datacurve.ai
Epoch's name: glm-5.2. Best closed model on this set: gpt-6-astra_xhigh, 74.1% (OpenAI). All 69 results
- effort max93.6%conceptualreasoning.ai
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5-2_unknown, 99.2% (Anthropic). All 241 results
- effort none; Harness: mini-swe-agent; Reasoning effort: none24.5%cognition.com
Epoch's name: glm-5.2. Best closed model on this set: claude-opus-5-5_medium, 54.6% (Anthropic). All 46 results
- effort unknown0.0%21 Jun 2026, gbaeval.com
Epoch's name: glm-5.2. Best closed model on this set: claude-opus-5-5_unknown, 90.6% (Anthropic). All 28 results
- effort max45.8%41.8 to 49.9conceptualreasoning.ai
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5-2_xhigh, 70.6% (Anthropic). All 203 results
- effort max; Scaffold: Claude Code30.4%via Epoch AI
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5-1_max, 44.6% (Anthropic). All 22 results
- effort max; Reasoning effort: max35.0%±4.8vals.ai
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5-1_max, 100.0% (Anthropic). All 85 results
- effort unknown58.8%lmcouncil.ai
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5-2_unknown, 94.9% (Anthropic). All 125 results
- effort high55.6%yhenon.github.io
Epoch's name: glm-5.2. Best closed model on this set: claude-fable-5_high, 95.0% (Anthropic). All 29 results
- effort max70.1%±0.0htihle.github.io
- effort high67.3%±0.0htihle.github.io
Epoch's name: glm-5.2. Best closed model on this set: gpt-6-astra_promax, 93.6% (OpenAI). All 173 results
Claimed by the maker Claimed
The model card has no claimed scores in a machine-readable form (the card's model-index), so none are shown. The maker may give scores in the card text: read the card.
Other views, which we link to and do not copy: look this model up on Artificial Analysis and the Arena leaderboard.
Independent scores from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Each row links its own source. What the badges mean.
Run it yourself
Memory needed = weights + context cache + 0.50 GB for the runtime. Context cache at 8k tokens: 0.69 GB, worked out from the model's config (zai-org/GLM-5.2), at 16-bit precision. Choose a machine above to see a verdict for each format. Method
GGUF for llama.cpp, Ollama, LM Studio and others
unsloth/GLM-5.2-GGUF (converted by unsloth), 13k downloads. Sizes are the real files on the Hub, all parts added up.
| Format | Size | Memory needed | Get it |
|---|---|---|---|
| UD-IQ3_XXS | 262.3 GBfile size, 7 parts | 263.5 GB262.3 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-IQ3_XXSDownload hf download unsloth/GLM-5.2-GGUF --include "UD-IQ3_XXS/GLM-5.2-UD-IQ3_XXS-*-of-00007.gguf" --local-dir GLM-5.2-GGUF |
| UD-IQ3_S | 287.4 GBfile size, 8 parts | 288.6 GB287.4 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-IQ3_SDownload hf download unsloth/GLM-5.2-GGUF --include "UD-IQ3_S/GLM-5.2-UD-IQ3_S-*-of-00008.gguf" --local-dir GLM-5.2-GGUF |
| UD-Q3_K_M | 319.2 GBfile size, 9 parts | 320.4 GB319.2 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-Q3_K_MDownload hf download unsloth/GLM-5.2-GGUF --include "UD-Q3_K_M/GLM-5.2-UD-Q3_K_M-*-of-00009.gguf" --local-dir GLM-5.2-GGUF |
| UD-Q3_K_XL | 319.4 GBfile size, 9 parts | 320.6 GB319.4 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-Q3_K_XLDownload hf download unsloth/GLM-5.2-GGUF --include "UD-Q3_K_XL/GLM-5.2-UD-Q3_K_XL-*-of-00009.gguf" --local-dir GLM-5.2-GGUF |
| UD-IQ4_XS | 340.2 GBfile size, 9 parts | 341.4 GB340.2 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-IQ4_XSDownload hf download unsloth/GLM-5.2-GGUF --include "UD-IQ4_XS/GLM-5.2-UD-IQ4_XS-*-of-00009.gguf" --local-dir GLM-5.2-GGUF |
| UD-IQ4_NL | 347.1 GBfile size, 9 parts | 348.3 GB347.1 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-IQ4_NLDownload hf download unsloth/GLM-5.2-GGUF --include "UD-IQ4_NL/GLM-5.2-UD-IQ4_NL-*-of-00009.gguf" --local-dir GLM-5.2-GGUF |
| UD-Q4_K_S | 406.5 GBfile size, 10 parts | 407.6 GB406.5 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-Q4_K_SDownload hf download unsloth/GLM-5.2-GGUF --include "UD-Q4_K_S/GLM-5.2-UD-Q4_K_S-*-of-00010.gguf" --local-dir GLM-5.2-GGUF |
| UD-Q4_K_M | 433.8 GBfile size, 11 parts | 435.0 GB433.8 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-Q4_K_MDownload hf download unsloth/GLM-5.2-GGUF --include "UD-Q4_K_M/GLM-5.2-UD-Q4_K_M-*-of-00011.gguf" --local-dir GLM-5.2-GGUF |
| UD-Q4_K_XL | 435.2 GBfile size, 11 parts | 436.4 GB435.2 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-Q4_K_XLDownload hf download unsloth/GLM-5.2-GGUF --include "UD-Q4_K_XL/GLM-5.2-UD-Q4_K_XL-*-of-00011.gguf" --local-dir GLM-5.2-GGUF |
| UD-Q5_K_S | 491.0 GBfile size, 12 parts | 492.2 GB491.0 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-Q5_K_SDownload hf download unsloth/GLM-5.2-GGUF --include "UD-Q5_K_S/GLM-5.2-UD-Q5_K_S-*-of-00012.gguf" --local-dir GLM-5.2-GGUF |
| UD-Q5_K_M | 522.3 GBfile size, 13 parts | 523.5 GB522.3 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-Q5_K_MDownload hf download unsloth/GLM-5.2-GGUF --include "UD-Q5_K_M/GLM-5.2-UD-Q5_K_M-*-of-00013.gguf" --local-dir GLM-5.2-GGUF |
| UD-Q5_K_XL | 523.8 GBfile size, 13 parts | 525.0 GB523.8 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-Q5_K_XLDownload hf download unsloth/GLM-5.2-GGUF --include "UD-Q5_K_XL/GLM-5.2-UD-Q5_K_XL-*-of-00013.gguf" --local-dir GLM-5.2-GGUF |
| UD-Q6_K | 582.9 GBfile size, 14 parts | 584.1 GB582.9 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-Q6_KDownload hf download unsloth/GLM-5.2-GGUF --include "UD-Q6_K/GLM-5.2-UD-Q6_K-*-of-00014.gguf" --local-dir GLM-5.2-GGUF |
| UD-Q6_K_XL | 637.4 GBfile size, 16 parts | 638.6 GB637.4 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-Q6_K_XLDownload hf download unsloth/GLM-5.2-GGUF --include "UD-Q6_K_XL/GLM-5.2-UD-Q6_K_XL-*-of-00016.gguf" --local-dir GLM-5.2-GGUF |
| Q8_0 | 746.3 GBfile size, 17 parts | 747.5 GB746.3 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:Q8_0Download hf download unsloth/GLM-5.2-GGUF --include "Q8_0/GLM-5.2-Q8_0-*-of-00017.gguf" --local-dir GLM-5.2-GGUF |
| UD-Q8_K_XL | 763.4 GBfile size, 18 parts | 764.6 GB763.4 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:UD-Q8_K_XLDownload hf download unsloth/GLM-5.2-GGUF --include "UD-Q8_K_XL/GLM-5.2-UD-Q8_K_XL-*-of-00018.gguf" --local-dir GLM-5.2-GGUF |
| BF16 | 1404.4 GBfile size, 33 parts | 1405.6 GB1404.4 + 0.69 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/GLM-5.2-GGUF:BF16Download hf download unsloth/GLM-5.2-GGUF --include "BF16/GLM-5.2-BF16-*-of-00033.gguf" --local-dir GLM-5.2-GGUF |
MLX for Apple silicon
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| MLXphilipjohnbasile/GLM-5.2-Demolition-q4a4-soul-MLX | 174.6 GBfile size | 175.8 GB174.6 + 0.69 + 0.50 | Download page |
| MLXmlx-community/GLM-5.2-DQ4plus-q8 | 432.7 GBfile size | 433.9 GB432.7 + 0.69 + 0.50 | Download page |
Original weights size from the safetensors header, for vLLM, SGLang, Transformers
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| BF16original weights | 1403.2 GBfile size | 1404.4 GB1403.2 + 0.69 + 0.50 |
hf download zai-org/GLM-5.2 --local-dir GLM-5.2Other conversions (18; sizes not read)
- zai-org/GLM-5.2-FP8 FP8, by the maker, 74k downloads
- nvidia/GLM-5.2-NVFP4 NVFP4, 30k downloads
- amd/GLM-5.2-MXFP4 Other, mxfp4, 18k downloads
- brandonmusic/GLM-5.2-EXL3-TR3v4-3.5bpw-MTP78 EXL3, 13k downloads
- malaiwah/GLM-5.2-EXL3-FQ-segments EXL3, 6.9k downloads
- mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp FP8, int4, 3.6k downloads
- local-inference-lab/GLM-5.2-NVFP4 NVFP4, 2.3k downloads
- RedHatAI/GLM-5.2-FP8 FP8, 1.1k downloads
- RedHatAI/GLM-5.2-NVFP4-FP8 NVFP4, 554 downloads
- QuantTrio/GLM-5.2-Int4-Int8Mix Other, int4, 392 downloads
- RedHatAI/GLM-5.2-NVFP4 NVFP4, 321 downloads
- mastouri/GLM-5.2-colibri-E8-IQ3-with-int8-mtp FP8, int8, 308 downloads
- amd/GLM-5.2-Quark-MXFP4-AttnFP8 Other, mxfp4, 296 downloads
- cyankiwi/GLM-5.2-AWQ-INT4 AWQ, int4, 268 downloads
- intel-ai/GLM-5.2-MXFP4-CT-AutoRound Other, mxfp4, 230 downloads
- canada-quant/GLM-5.2-W4A16-MTP GPTQ, w4a16, 101 downloads
- jpsequeira/GLM-5.2-EXL3-TR3-3.40bpw-KVarN-K4V2 EXL3, 49 downloads
- willfalco/GLM-5.2-EXL3-TR3-3.42bpw EXL3, 38 downloads
The Hub tags 100 or more quantised repos as conversions of this model; we read the most downloaded and show the maker's own first.
Run it by API
18 providers serve this model through the Hugging Face router. Prices are US dollars per million tokens. Precision is what the provider reports; a lower precision than the original weights can change answers. Live first, then cheapest output.
| Provider | In | Out | Context | Precision | Tokens/s | First token | Uptime 24h | Tools | JSON |
|---|---|---|---|---|---|---|---|---|---|
| DeepInfra | $0.56 | $1.80 | 1M | fp4 | 54 | 1.9 s | 97.9% | yes | yes |
| Morph | $0.64 | $1.98 | 1M | fp8 | 82 | 0.7 s | 100.0% | yes | yes |
| Novita | $0.65 | $2.04 | 1M | fp8 | 48 | 2.3 s | 100.0% | yes | no |
| SiliconFlow | $0.65 | $2.05 | 1M | fp8 | 70 | 1.2 s | 100.0% | yes | no |
| DigitalOcean | $0.70 | $2.20 | 1M | not given | 75 | 0.7 s | 99.7% | yes | yes |
| Relace | $0.31 | $2.40 | 1M | fp4 | 102 | 0.8 s | 99.9% | yes | no |
| CoreWeave | $0.62 | $2.42 | 1M | fp4 | 141 | 0.8 s | 100.0% | yes | yes |
| BaseTen | $1.40 | $4.40 | 1M | fp8 | 57 | 1.9 s | 100.0% | yes | yes |
| Mistral | $1.40 | $4.40 | 1M | not given | 99 | 0.9 s | 100.0% | yes | yes |
| Nebius | $1.40 | $4.40 | 1M | fp4 | ? | ? | 40.5% | yes | yes |
| Parasail | $1.40 | $4.40 | 256k | fp4 | 123 | 0.6 s | 100.0% | yes | yes |
| Together | $1.40 | $4.40 | 1024k | not given | 67 | 0.5 s | 99.8% | yes | yes |
| Venice | $1.40 | $4.40 | 977k | fp8 | 99 | 1.6 s | 99.8% | yes | yes |
| Z.AIthe maker | $1.40 | $4.40 | 1M | fp8 | 47 | 1.8 s | 100.0% | yes | no |
| Wafer | $0.007 | $6.80 | 1M | fp4 | 122 | 0.5 s | 100.0% | yes | yes |
| Cloudflare | $0.19 | $8.00 | 256k | not given | 57 | 0.9 s | 100.0% | yes | yes |
| featherless-ai | ? | ? | ? | not given | ? | ? | ? | not given | not given |
| zai-orgthe makerserves zai-org/GLM-5.2-FP8 | ? | ? | ? | not given | 40 | 1.4 s | ? | yes | no |
Through the router, with a Hugging Face token (the router bills your Hugging Face account). Change the name after the colon to pick another provider.
curl https://router.huggingface.co/v1/chat/completions \
-H "Authorization: Bearer $HF_TOKEN" -H "Content-Type: application/json" \
-d '{"model": "zai-org/GLM-5.2:deepinfra", "messages": [{"role": "user", "content": "Hello"}]}'Not built yet: one key for every open model, with BlockRun, and our own tested-quality score per provider. Until then, precision and measured speed are the best guide to a provider here.
Planned for this page Not built yet
- Try it in private chat
- A link to Solana AI's confidential chat for the models it serves. This model is not listed as served there.
- Fine-tune it
- Fine-tuning a few base models from this page and publishing the result here.
- Creators and earnings
- Pages for people who publish fine-tunes and adapters of this model, and what they earn.
- Fund and bounties
- Bounties for new open models, paid from a public fund. How the fund will work.
- Contribute compute
- How to give compute to open model work, with Akash.
Model data from the Hugging Face Hub 19 Nov 2026, 13:43 UTC. The repo. Found a mistake on this page? Tell us.