- Size
- 321B 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
- 25 Aug 2026the date the repo was created
- Context length
- 1M tokensfrom the model's config
- Kind
- Base model (no base model declared on the Hub)Glm5NextForConditionalGeneration
- On Hugging Face
- zai-org/GLM-5.3-Flash6.4M downloads in the last 30 days, 3.3k likes
Scores
- effort max21.0%±4.113 Oct 2026, log
Epoch's name: glm-5.3-flash. Best closed model on this set: claude-fable-5-2_max, 79.0% (Anthropic). All 166 results
- effort max17.1%±5.927 Aug 2026, Epoch AI
Epoch's name: glm-5.3-flash. Best closed model on this set: claude-fable-5-2_max, 100.0% (Anthropic). All 83 results
- effort max55.8%±2.927 Aug 2026, Epoch AI
Epoch's name: glm-5.3-flash. Best closed model on this set: claude-fable-5-2_max, 95.1% (Anthropic). All 127 results
- effort max90.2%±1.726 Aug 2026, log
Epoch's name: glm-5.3-flash. Best closed model on this set: gpt-6.2-astra_max, 96.8% (OpenAI). All 341 results
- effort max8.0%±2.728 Aug 2026, Epoch AI
Epoch's name: glm-5.3-flash. Best closed model on this set: gpt-6.2-astra_max, 95.0% (OpenAI). All 148 results
- effort max93.9%±2.226 Aug 2026, log
Epoch's name: glm-5.3-flash. Best closed model on this set: claude-fable-5_high, 100.0% (Anthropic). All 318 results
- effort unknown52.8%via Epoch AI (error bar left out)
Epoch's name: glm-5.3-flash. Best closed model on this set: gemini-4-carbon_unknown, 84.2% (Google DeepMind). All 66 results
- effort max91.0%arcprize.org
- effort high71.8%arcprize.org
- effort low47.0%arcprize.org
Epoch's name: glm-5.3-flash. Best closed model on this set: claude-fable-5_max, 98.5% (Anthropic). All 318 results
- effort max65.8%via Epoch AI
- effort high50.1%via Epoch AI
- effort low27.9%via Epoch AI
Epoch's name: glm-5.3-flash. Best closed model on this set: gpt-6-astra_max, 95.0% (OpenAI). All 299 results
- effort unknown15.4%via Epoch AI
Epoch's name: glm-5.3-flash. Best closed model on this set: gemini-4-carbon_high, 36.0% (Google DeepMind). All 271 results
- effort max; Reasoning level: Max36.8%cursor.com
- effort high; Reasoning level: High31.1%cursor.com
- effort low; Reasoning level: Low26.9%cursor.com
Epoch's name: glm-5.3-flash. Best closed model on this set: gemini-4-carbon_high, 60.4% (Google DeepMind). All 94 results
- effort max; Harness: mini-swe-agent; Reasoning effort: max63.4%59.0 to 67.8deepswe.datacurve.ai
Epoch's name: glm-5.3-flash. Best closed model on this set: gpt-6-astra_xhigh, 74.1% (OpenAI). All 69 results
- effort max; Harness: devin-cli; Reasoning effort: max31.8%cognition.com
Epoch's name: glm-5.3-flash. Best closed model on this set: claude-opus-5-5_medium, 54.6% (Anthropic). All 46 results
- effort max; Harness: proximus18.1%frontierswe.com
Epoch's name: glm-5.3-flash. Best closed model on this set: claude-fable-5-2_max, 69.9% (Anthropic). All 25 results
- effort max; Scaffold: Claude Code36.8%via Epoch AI
Epoch's name: glm-5.3-flash. Best closed model on this set: claude-fable-5-1_max, 44.6% (Anthropic). All 22 results
- effort max21.0%±4.1vals.ai
Epoch's name: glm-5.3-flash. Best closed model on this set: claude-fable-5-1_max, 100.0% (Anthropic). All 85 results
- effort max52.5%yhenon.github.io
Epoch's name: glm-5.3-flash. Best closed model on this set: claude-fable-5_high, 95.0% (Anthropic). All 29 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: 1.4 GB, worked out from the model's config (zai-org/GLM-5.3-Flash), 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
peasantsmith/GLM-5.3-Flash-Maya-GGUF (converted by peasantsmith), 1.5M downloads. Sizes are the real files on the Hub, all parts added up. For image input you also need the projector file vision/mmproj-GLM-5.3-Flash-F16.gguf (1.1 GB), not counted below.
| Format | Size | Memory needed | Get it |
|---|---|---|---|
| IQ1_S | 68.4 GBfile size, 2 parts | 70.2 GB68.4 + 1.4 + 0.50 | Commandsllama.cpp llama-server -hf peasantsmith/GLM-5.3-Flash-Maya-GGUF:IQ1_SDownload hf download peasantsmith/GLM-5.3-Flash-Maya-GGUF --include "Maya-XS/GLM-5.3-Flash-Maya-XS-IQ1_S-*-of-00002.gguf" --local-dir GLM-5.3-Flash-Maya-GGUF |
| IQ2_XXS_S | 88.2 GBfile size, 3 parts | 90.1 GB88.2 + 1.4 + 0.50 | Commandsllama.cpp llama-server -hf peasantsmith/GLM-5.3-Flash-Maya-GGUF:IQ2_XXS_SDownload hf download peasantsmith/GLM-5.3-Flash-Maya-GGUF --include "Maya-S24/GLM-5.3-Flash-Maya-S24-IQ2_XXS_S-*-of-00003.gguf" --local-dir GLM-5.3-Flash-Maya-GGUF |
| IQ2_XXS | 89.8 GBfile size, 3 parts | 91.7 GB89.8 + 1.4 + 0.50 | Commandsllama.cpp llama-server -hf peasantsmith/GLM-5.3-Flash-Maya-GGUF:IQ2_XXSDownload hf download peasantsmith/GLM-5.3-Flash-Maya-GGUF --include "Maya-S-v2-IQ2_XXS/GLM-5.3-Flash-Maya-S-v2-IQ2_XXS-*-of-00003.gguf" --local-dir GLM-5.3-Flash-Maya-GGUF |
| IQ2_S | 108.0 GBfile size, 3 parts | 109.9 GB108.0 + 1.4 + 0.50 | Commandsllama.cpp llama-server -hf peasantsmith/GLM-5.3-Flash-Maya-GGUF:IQ2_SDownload hf download peasantsmith/GLM-5.3-Flash-Maya-GGUF --include "Maya-M/GLM-5.3-Flash-Maya-M-IQ2_S-*-of-00003.gguf" --local-dir GLM-5.3-Flash-Maya-GGUF |
| IQ3_S | 145.6 GBfile size, 4 parts | 147.5 GB145.6 + 1.4 + 0.50 | Commandsllama.cpp llama-server -hf peasantsmith/GLM-5.3-Flash-Maya-GGUF:IQ3_SDownload hf download peasantsmith/GLM-5.3-Flash-Maya-GGUF --include "Maya-L/GLM-5.3-Flash-Maya-L-IQ3_S-*-of-00004.gguf" --local-dir GLM-5.3-Flash-Maya-GGUF |
MLX for Apple silicon
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| MLXJundot/GLM-5.3-Flash-oQ4e-mtp | 173.6 GBfile size | 175.5 GB173.6 + 1.4 + 0.50 | Download page |
| MLXorcarouter/GLM-5.3-Flash-MLX | 1057.3 GBfile size | 1059.2 GB1057.3 + 1.4 + 0.50 | Download page |
Original weights size from the safetensors header, for vLLM, SGLang, Transformers
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| F8_E4M3original weights | 305.7 GBfile size | 307.6 GB305.7 + 1.4 + 0.50 |
hf download zai-org/GLM-5.3-Flash --local-dir GLM-5.3-FlashOther conversions (17; sizes not read)
- zai-org/GLM-5.3-Flash-BF16 Other, bf16, by the maker, 24k downloads
- tattrongvu/GLM-5.3-Flash-NVFP4 NVFP4, 90k downloads
- nvidia/GLM-5.3-Flash-NVFP4 NVFP4, 51k downloads
- INCModel3/GLM-5.3-Flash-MXFP4-Mixed-CT-AutoRound Other, mxfp4, 22k downloads
- autotrust/GLM5.3-Flash-E192-DGX-Spark NVFP4, 22k downloads
- RedHatAI/GLM-5.3-Flash-NVFP4 NVFP4, 18k downloads
- amd/GLM-5.3-Flash-Quark-MXFP4 Other, mxfp4, 15k downloads
- nvidia/GLM-5.3-Flash-FP4-Local Other, 14k downloads
- Mia-AiLab/GLM-5.3-Flash-EXL3-4bpw-TensorFold EXL3, 4.4k downloads
- wtdcode/GLM-5.3-Flash-AWQ-W4A16 AWQ, w4a16, 2.6k downloads
- canada-quant/GLM-5.3-Flash-W4A16-MTP GPTQ, w4a16, 2.6k downloads
- turboderp/GLM-5.3-Flash-exl3 EXL3, 2.3k downloads
- beamster/GLM-5.3-Flash-Sushi-2.4bpw EXL3, 1.0k downloads
- unsloth/GLM-5.3-Flash-FP8 FP8, 852 downloads
- GRRNMAKE/GLM5.3Flash_symb FP8, 847 downloads
- cyankiwi/GLM-5.3-Flash-AWQ-INT4 AWQ, int4, 585 downloads
- yamz-labs/GLM-5.3-Flash-EXL3-Yamz EXL3, 509 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
31 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 |
|---|---|---|---|---|---|---|---|---|---|
| Relace | $0.04 | $0.078 | 1M | fp4 | 84 | 1.0 s | 99.9% | yes | no |
| Wafer | $0.042 | $0.18 | 1M | nvfp4 | 44 | 0.7 s | 100.0% | yes | yes |
| Makora | $0.14 | $0.24 | 1M | fp4 | 94 | 0.5 s | 99.4% | yes | yes |
| Tenstorrent | $0.075 | $0.25 | 256k | fp8 | 37 | 3.6 s | 100.0% | yes | yes |
| Novita | $0.084 | $0.28 | 1M | fp8 | 40 | 2.0 s | 100.0% | yes | yes |
| Morph | $0.10 | $0.30 | 1M | fp8 | 103 | 0.6 s | 99.8% | yes | yes |
| Reka | $0.12 | $0.30 | 1M | not given | 101 | 0.5 s | 100.0% | yes | yes |
| InferenceNet | $0.11 | $0.35 | 1M | fp4 | 138 | 0.4 s | 100.0% | yes | yes |
| DigitalOcean | $0.11 | $0.38 | 1M | not given | 80 | 1.0 s | 100.0% | yes | yes |
| CoreWeave | $0.12 | $0.38 | 1M | nvfp4 | 150 | 0.6 s | 99.0% | yes | yes |
| River | $0.12 | $0.40 | 1M | not given | 56 | 1.2 s | 99.7% | yes | yes |
| OpenInference | $0.05 | $0.45 | 1M | fp4 | 137 | 0.8 s | 100.0% | yes | yes |
| BaseTen | $0.15 | $0.50 | 1M | fp8 | 64 | 0.5 s | 99.9% | yes | yes |
| Cohere | $0.15 | $0.50 | 1M | nvfp4 | 145 | 0.7 s | 100.0% | yes | yes |
| Crusoe | $0.15 | $0.50 | 1M | fp4 | 96 | 0.7 s | 100.0% | yes | yes |
| DeepInfra | $0.15 | $0.50 | 1M | fp4 | 59 | 1.2 s | 100.0% | yes | yes |
| DekaLLM | $0.10 | $0.50 | 1M | not given | 31 | 2.5 s | 99.3% | yes | yes |
| Fireworks | $0.15 | $0.50 | 1M | not given | 80 | 1.6 s | 100.0% | yes | yes |
| Inceptron | $0.15 | $0.50 | 1M | fp8 | 79 | 0.5 s | 100.0% | yes | yes |
| Mistral | $0.15 | $0.50 | 1M | not given | 142 | 0.9 s | 100.0% | yes | yes |
| Modal | $0.15 | $0.50 | 1M | fp4 | 110 | 0.4 s | 100.0% | yes | yes |
| Near AI | $0.08 | $0.50 | 1M | fp8 | 17 | 1.7 s | 99.9% | yes | yes |
| Nebius | $0.15 | $0.50 | 1M | fp4 | 72 | 0.7 s | 100.0% | yes | yes |
| Parasail | $0.15 | $0.50 | 1M | fp8 | 49 | 1.7 s | 100.0% | yes | yes |
| Particle AI | $0.13 | $0.50 | 1M | nvfp4 | 98 | 0.7 s | 100.0% | yes | yes |
| Together | $0.15 | $0.50 | 1024k | not given | 139 | 0.5 s | 100.0% | yes | yes |
| Venice | $0.15 | $0.50 | 1M | not given | 65 | 0.7 s | 100.0% | yes | yes |
| Z.AIthe maker | $0.15 | $0.50 | 1M | fp8 | 46 | 1.8 s | 100.0% | yes | yes |
| Sail Research | $0.17 | $0.63 | 1M | fp4 | 47 | 2.6 s | 100.0% | yes | yes |
| featherless-ai | ? | ? | ? | not given | ? | ? | ? | not given | not given |
| zai-orgthe makerserves zai-org/GLM-5.3-Flash-BF16 | ? | ? | ? | not given | 45 | 1.3 s | ? | yes | yes |
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-5.3-Flash:relace", "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.
Fine-tunes of this model
Repos that declare this model as their base, most downloaded first. Listing is not a recommendation; unreviewed ones have not been checked.
- orcarouter/GLM-5.3-Flash-Uncensored-FP8321B11kUnreviewed
- incoai/GLM-5.3-Flash-DFlash21.2B11kUnreviewed
- inference-optimization/GLM-5.3-Flash-0.1B-A0.1B84M8.2kUnreviewed
- 0xSero/GLM-5.3-Flash-DSpine1.2B833Unreviewed
- ukisai/Swift1.5-GLM-5.3-Flash321B387Unreviewed
- paperinstruments/Ultramarine-1-Flash321B195Unreviewed
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.