- Size
- 32.8B parameterscounted from the safetensors header on the Hub
- Licence
- custom, read it, Apache License 2.0Our reading, check the licence: A custom licence we have not read for this model: other makers use the same name for different texts. Read the model's own licence file before you use it.Licence text
- On the Hub since
- 5 Mar 2025the date the repo was created
- Context length
- 40k tokensfrom the model's config
- Fine-tune
- of Qwen/Qwen2.5-32BQwen2ForCausalLM
- On Hugging Face
- Qwen/QwQ-32B30k downloads in the last 30 days, 3.0k likes
Scores
- default setting65.3%±2.828 Aug 2026, log
Epoch's name: QwQ-32B. Best closed model on this set: gpt-6.2-astra_max, 96.8% (OpenAI). All 341 results
- default setting59.2%±6.328 Aug 2026, log
Epoch's name: QwQ-32B. Best closed model on this set: claude-fable-5_high, 100.0% (Anthropic). All 318 results
- Edit format: diff20.9%6 Mar 2025, aider.chat
Epoch's name: QwQ-32B. Best closed model on this set: gpt-5-2025-08-07_high, 88.0% (OpenAI). All 72 results
- default setting83.3%Fiction.live leaderboard
Epoch's name: chutes/QwQ-32B. Best closed model on this set: gpt-5-2025-08-07_medium, 97.2% (OpenAI). All 61 results
- default setting80.2%lechmazur/writing Github repository
Epoch's name: QwQ-32B (16K thinking). Best closed model on this set: gpt-5-2025-08-07_medium, 86.0% (OpenAI). All 49 results
- default setting72.0%LiveBench Leaderboard
Epoch's name: QwQ-32B. Best closed model on this set: gemini-2.5-pro-exp-03-25, 82.3% (Google DeepMind). All 54 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: 2.0 GB, worked out from the model's config (Qwen/QwQ-32B), 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
Qwen/QwQ-32B-GGUF (the maker's own conversion), 20k downloads. Sizes are the real files on the Hub, all parts added up.
| Format | Size | Memory needed | Get it |
|---|---|---|---|
| Q4_K_M | 18.5 GBfile size | 21.0 GB18.5 + 2.0 + 0.50 | Commandsllama.cpp llama-server -hf Qwen/QwQ-32B-GGUF:Q4_K_MOllama ollama run hf.co/Qwen/QwQ-32B-GGUF:Q4_K_MDownload hf download Qwen/QwQ-32B-GGUF --include "qwq-32b-q4_k_m.gguf" --local-dir QwQ-32B-GGUF |
| Q5_0 | 21.1 GBfile size | 23.6 GB21.1 + 2.0 + 0.50 | Commandsllama.cpp llama-server -hf Qwen/QwQ-32B-GGUF:Q5_0Ollama ollama run hf.co/Qwen/QwQ-32B-GGUF:Q5_0Download hf download Qwen/QwQ-32B-GGUF --include "qwq-32b-q5_0.gguf" --local-dir QwQ-32B-GGUF |
| Q5_K_M | 21.7 GBfile size | 24.2 GB21.7 + 2.0 + 0.50 | Commandsllama.cpp llama-server -hf Qwen/QwQ-32B-GGUF:Q5_K_MOllama ollama run hf.co/Qwen/QwQ-32B-GGUF:Q5_K_MDownload hf download Qwen/QwQ-32B-GGUF --include "qwq-32b-q5_k_m.gguf" --local-dir QwQ-32B-GGUF |
| Q6_K | 25.0 GBfile size | 27.5 GB25.0 + 2.0 + 0.50 | Commandsllama.cpp llama-server -hf Qwen/QwQ-32B-GGUF:Q6_KOllama ollama run hf.co/Qwen/QwQ-32B-GGUF:Q6_KDownload hf download Qwen/QwQ-32B-GGUF --include "qwq-32b-q6_k.gguf" --local-dir QwQ-32B-GGUF |
| Q8_0 | 32.4 GBfile size | 34.9 GB32.4 + 2.0 + 0.50 | Commandsllama.cpp llama-server -hf Qwen/QwQ-32B-GGUF:Q8_0Ollama ollama run hf.co/Qwen/QwQ-32B-GGUF:Q8_0Download hf download Qwen/QwQ-32B-GGUF --include "qwq-32b-q8_0.gguf" --local-dir QwQ-32B-GGUF |
| F16 | 61.0 GBfile size, 17 parts | 63.5 GB61.0 + 2.0 + 0.50 | Commandsllama.cpp llama-server -hf Qwen/QwQ-32B-GGUF:F16Download hf download Qwen/QwQ-32B-GGUF --include "fp16/qwq-32b-fp16-*-of-00017.gguf" --local-dir QwQ-32B-GGUF |
MLX for Apple silicon
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| MLX 4bitmlx-community/QwQ-32B-4bit | 17.2 GBfile size | 19.7 GB17.2 + 2.0 + 0.50 | Download page |
| MLX 8bitlmstudio-community/QwQ-32B-MLX-8bit | 32.4 GBfile size | 34.9 GB32.4 + 2.0 + 0.50 | Download page |
Original weights size from the safetensors header, for vLLM, SGLang, Transformers
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| BF16original weights | 61.0 GBfile size | 63.5 GB61.0 + 2.0 + 0.50 |
hf download Qwen/QwQ-32B --local-dir QwQ-32BOther conversions (23; sizes not read)
- Qwen/QwQ-32B-AWQ AWQ, by the maker, 7.3k downloads
- RedHatAI/QwQ-32B-FP8-dynamic FP8, 221 downloads
- unsloth/QwQ-32B-unsloth-bnb-4bit bitsandbytes, 4bit, 209 downloads
- unsloth/QwQ-32B-bnb-4bit bitsandbytes, 4bit, 192 downloads
- fantos/QwQ-32B-bnb-4bit bitsandbytes, 4bit, 50 downloads
- ginipick/QwQ-32B-NF4 bitsandbytes, 44 downloads
- drawais/QwQ-32B-AWQ-INT4 AWQ, int4, 37 downloads
- ModelCloud/QwQ-32B-gptqmodel-4bit-vortex-v1 GPTQ, 4bit, 21 downloads
- drawais/QwQ-32B-NVFP4 AWQ, nvfp4, 19 downloads
- mlc-ai/QwQ-32B-q4f16_1-MLC Other, 15 downloads
- bartowski/QwQ-32B-exl2 EXL2, 12 downloads
- OPEA/QwQ-32B-int4-AutoRound-awq-asym AWQ, int4, 12 downloads
- Firworks/QwQ-32B-nvfp4 NVFP4, 12 downloads
- mlc-ai/QwQ-32B-q4f16_0-MLC Other, 11 downloads
- cortecs/QwQ-32B-FP8-Dynamic FP8, 11 downloads
- qingcheng-ai/QWQ-32B-FP8 FP8, 11 downloads
- cshared/Qwen-QwQ-32B-5.0bpw-exl2 EXL2, 10 downloads
- cshared/Qwen-QwQ-32B-5.5bpw-exl2 EXL2, 9 downloads
- Jellon/QwQ-32B-exl3-4bpw EXL3, 9 downloads
- celertor-ai/QWQ-32B-FP8 FP8, 9 downloads
- dillonroach/Qwen-QwQ-32B-425bpw-h6-exl2 EXL2, 8 downloads
- kaitchup/QwQ-32B-AutoRoundGPTQ-4bit GPTQ, 4bit, 8 downloads
- kaitchup/QwQ-32B-AutoRoundGPTQ-2bit GPTQ, 2bit, 8 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
No provider on Hugging Face Inference Providers serves this model (as of 19 Nov 2026, 13:58 UTC). Other hosts may; we only list what the router reports.
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.