- Size
- 27.8B parameterscounted from the safetensors header on the Hub
- Licence
- unknown, Apache License 2.0Our reading, check the licence: You can use, change and redistribute the model, including commercially. Keep the licence and any notices, and say which files you changed; you do not have to publish your changes. It includes a patent licence that ends for anyone who sues over patents in the work.
- On the Hub since
- 5 Aug 2026the date the repo was created
- Context length
- 256k tokensfrom the model's config
- Kind
- Base model (no base model declared on the Hub)Qwen3_5ForConditionalGeneration
- On Hugging Face
- Qwen/Qwen3.8-27B6.6M downloads in the last 30 days, 21k likes
Scores
- default setting47.5%via Epoch AI (error bar left out)
Epoch's name: qwen3.8-27b. Best closed model on this set: gemini-4-carbon_unknown, 84.2% (Google DeepMind). All 66 results
- effort xhigh87.5%arcprize.org
- effort low69.2%arcprize.org
- effort medium68.7%arcprize.org
- effort none34.0%arcprize.org
Epoch's name: qwen3.8-27b. Best closed model on this set: claude-fable-5_max, 98.5% (Anthropic). All 318 results
- effort xhigh42.4%via Epoch AI
- effort low22.8%via Epoch AI
- effort medium13.2%via Epoch AI
- effort none1.5%via Epoch AI
Epoch's name: qwen3.8-27b. Best closed model on this set: gpt-6-astra_max, 95.0% (OpenAI). All 299 results
- effort xhigh5.4%via Epoch AI
- effort none0.3%via Epoch AI
- effort low0.0%via Epoch AI
- effort medium0.0%via Epoch AI
Epoch's name: qwen3.8-27b. Best closed model on this set: gemini-4-carbon_high, 36.0% (Google DeepMind). All 271 results
- default setting88.0%conceptualreasoning.ai
Epoch's name: qwen3.8-27b. Best closed model on this set: claude-fable-5-2_unknown, 99.2% (Anthropic). All 241 results
- Tools: None38.0%geobench.org
Epoch's name: qwen3.8-27b. Best closed model on this set: claude-fable-5_high, 89.0% (Anthropic). All 51 results
- default setting41.4%37.1 to 45.6conceptualreasoning.ai
Epoch's name: qwen3.8-27b. Best closed model on this set: claude-fable-5-2_xhigh, 70.6% (Anthropic). All 203 results
- default setting16.0%vals.ai
Epoch's name: qwen3.8-27b. Best closed model on this set: claude-fable-5-1_max, 100.0% (Anthropic). All 85 results
- default setting60.2%simple-bench.com
Epoch's name: qwen3.8-27b. Best closed model on this set: claude-fable-5-2_unknown, 94.9% (Anthropic). All 125 results
- effort xhigh45.0%yhenon.github.io
Epoch's name: qwen3.8-27b. 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: 0.50 GB, worked out from the model's config (Qwen/Qwen3.8-27B), 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/Qwen3.8-27B-GGUF (converted by unsloth), 7.1M downloads. Sizes are the real files on the Hub, all parts added up. For image input you also need the projector file mmproj-F16.gguf (0.86 GB), not counted below.
| Format | Size | Memory needed | Get it |
|---|---|---|---|
| UD-IQ3_XXS | 10.2 GBfile size | 11.2 GB10.2 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-IQ3_XXSOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-IQ3_XXSDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-IQ3_XXS.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-IQ3_S | 11.2 GBfile size | 12.2 GB11.2 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-IQ3_SOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-IQ3_SDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-IQ3_S.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q3_K_XL | 12.2 GBfile size | 13.2 GB12.2 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q3_K_XLOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q3_K_XLDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q3_K_XL.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-IQ4_XS | 13.4 GBfile size | 14.4 GB13.4 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-IQ4_XSOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-IQ4_XSDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-IQ4_XS.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q4_K_S | 14.3 GBfile size | 15.3 GB14.3 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q4_K_SOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q4_K_SDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q4_K_S.gguf" --local-dir Qwen3.8-27B-GGUF |
| Q4_0 | 15.0 GBfile size | 16.0 GB15.0 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:Q4_0Ollama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:Q4_0Download hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-Q4_0.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q4_K_M | 15.4 GBfile size | 16.4 GB15.4 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q4_K_MOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q4_K_MDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q4_K_M.gguf" --local-dir Qwen3.8-27B-GGUF |
| Q4_1 | 16.3 GBfile size | 17.3 GB16.3 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:Q4_1Ollama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:Q4_1Download hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-Q4_1.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q4_K_XL | 16.4 GBfile size | 17.4 GB16.4 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q4_K_XLOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q4_K_XLDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q4_K_XL.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q5_K_S | 17.5 GBfile size | 18.5 GB17.5 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_SOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_SDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q5_K_S.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q5_K_M | 18.5 GBfile size | 19.5 GB18.5 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_MOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_MDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q5_K_M.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q5_K_XL | 19.5 GBfile size | 20.5 GB19.5 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_XLOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_XLDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q5_K_XL.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q6_K | 20.5 GBfile size | 21.5 GB20.5 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q6_KOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q6_KDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q6_K.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q6_K_M | 21.5 GBfile size | 22.5 GB21.5 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q6_K_MOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q6_K_MDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q6_K_M.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q6_K_L | 22.5 GBfile size | 23.5 GB22.5 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q6_K_LOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q6_K_LDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q6_K_L.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q6_K_XL | 23.6 GBfile size | 24.6 GB23.6 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q6_K_XLOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q6_K_XLDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q6_K_XL.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q8_K_L | 26.1 GBfile size | 27.1 GB26.1 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q8_K_LOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q8_K_LDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q8_K_L.gguf" --local-dir Qwen3.8-27B-GGUF |
| Q8_0 | 27.1 GBfile size | 28.1 GB27.1 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:Q8_0Ollama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:Q8_0Download hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-Q8_0.gguf" --local-dir Qwen3.8-27B-GGUF |
| UD-Q8_K_XL | 29.3 GBfile size | 30.3 GB29.3 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q8_K_XLOllama ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q8_K_XLDownload hf download unsloth/Qwen3.8-27B-GGUF --include "Qwen3.8-27B-UD-Q8_K_XL.gguf" --local-dir Qwen3.8-27B-GGUF |
| BF16 | 50.9 GBfile size, 2 parts | 51.9 GB50.9 + 0.50 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Qwen3.8-27B-GGUF:BF16Download hf download unsloth/Qwen3.8-27B-GGUF --include "BF16/Qwen3.8-27B-BF16-*-of-00002.gguf" --local-dir Qwen3.8-27B-GGUF |
MLX for Apple silicon
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| MLX 4bitlmstudio-community/Qwen3.8-27B-MLX-4bit | 15.0 GBfile size | 16.0 GB15.0 + 0.50 + 0.50 | Download page |
| MLX 8bitlmstudio-community/Qwen3.8-27B-MLX-8bit | 27.5 GBfile size | 28.5 GB27.5 + 0.50 + 0.50 | Download page |
Original weights size from the safetensors header, for vLLM, SGLang, Transformers
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| BF16original weights | 51.7 GBfile size | 52.7 GB51.7 + 0.50 + 0.50 |
hf download Qwen/Qwen3.8-27B --local-dir Qwen3.8-27BOther conversions (19; sizes not read)
- Qwen/Qwen3.8-27B-FP8 FP8, by the maker, 6.4M downloads
- cdiamond/Qwen3.8-27B-iMatrix-NVFP4-MTP-GGUF GGUF, nvfp4, 3.3M downloads
- unsloth/Qwen3.8-27B-NVFP4 NVFP4, 1.8M downloads
- cyankiwi/Qwen3.8-27B-AWQ-INT4 AWQ, int4, 892k downloads
- RadixArk/Qwen3.8-27B-NVFP4 NVFP4, 635k downloads
- nicosuter/Qwen3.8-27B-AWQ AWQ, 438k downloads
- gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090 NVFP4, 299k downloads
- nvidia/Qwen3.8-27B-NVFP4 NVFP4, 297k downloads
- philbert440/Qwen3.8-27B-W4A16-AWQ AWQ, w4a16, 231k downloads
- RedHatAI/Qwen3.8-27B-INT4 Other, int4, 230k downloads
- SergiioB/Qwen3.8-27B-GPTQ-Int4-sym-G128-MTP-BF16 GPTQ, int4, 136k downloads
- Ar4ikov/Qwen3.8-27B-AWQ-W4A16-ASYM AWQ, w4a16, 73k downloads
- sokada4/Qwen3.8-27B-GPTQ-Int4 GPTQ, int4, 61k downloads
- unsloth/Qwen3.8-27B-unsloth-bnb-4bit bitsandbytes, 4bit, 59k downloads
- neroued/Qwen3.8-27B-NInfer Other, 58k downloads
- WarlaxZ/Qwen3.8-27B-NInfer-3090 Other, 56k downloads
- mikeinnyc/Qwen3.8-27B-GPTQ-Int4-sym-G128-MTP-BF16 GPTQ, int4, 25k downloads
- pearsonkyle/Qwen3.8-27B-GPTQ-W4A16 GPTQ, w4a16, 22k downloads
- intel-ai/Qwen3.8-27B-MXFP8-MXFP8MTP-CT-AutoRound Other, 14k 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
23 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 |
|---|---|---|---|---|---|---|---|---|---|
| Near AI | $0.04 | $1.35 | 256k | fp8 | 73 | 1.4 s | 99.9% | yes | yes |
| Cerebras | $0.99 | $1.49 | 64k | fp16 | 413 | 0.4 s | 99.9% | yes | no |
| Mancer 2 | $0.13 | $1.50 | 256k | fp8 | 69 | 0.9 s | 88.9% | yes | yes |
| Tenstorrent | $0.15 | $1.60 | 256k | fp8 | 100 | 0.7 s | 100.0% | yes | yes |
| DeepInfra | $0.15 | $1.88 | 256k | fp8 | 36 | 0.7 s | 99.9% | yes | yes |
| Ionstream | $0.04 | $1.95 | 256k | fp8 | 95 | 0.4 s | 99.2% | yes | yes |
| AkashML | $0.23 | $1.98 | 256k | fp8 | 36 | 0.9 s | 99.4% | yes | yes |
| Parasail | $0.24 | $2.20 | 256k | fp8 | 40 | 0.6 s | 100.0% | yes | yes |
| Wafer | $0.06 | $2.20 | 256k | fp8 | 239 | 0.4 s | 100.0% | yes | yes |
| Modal | $0.04 | $2.30 | 256k | not given | 109 | 0.4 s | 100.0% | yes | yes |
| DekaLLM | $0.04 | $2.50 | 256k | not given | 117 | 0.8 s | 100.0% | yes | yes |
| ModelRun | $0.20 | $2.50 | 256k | fp4 | 200 | 0.2 s | 100.0% | yes | yes |
| nscale | $0.20 | $2.50 | 1M | not given | 117 | 0.4 s | ? | yes | yes |
| Reka | $0.15 | $2.50 | 256k | fp8 | 74 | 0.7 s | 100.0% | yes | yes |
| SiliconFlow | $0.42 | $2.55 | 256k | fp8 | 41 | 1.0 s | 100.0% | yes | no |
| PrimeIntellect | $0.36 | $2.57 | 256k | fp8 | 8 | 2.9 s | 100.0% | yes | yes |
| CoreWeave | $0.40 | $3.00 | 256k | fp8 | 89 | 0.5 s | 99.6% | yes | yes |
| Nebius | $0.45 | $3.00 | 977k | fp4 | 209 | 0.8 s | 100.0% | yes | yes |
| Novita | $0.42 | $3.00 | 977k | not given | 44 | 0.8 s | 100.0% | yes | no |
| ovhcloud | $0.47 | $3.19 | 256k | not given | 28 | 0.6 s | ? | yes | yes |
| Cloudflare | $0.45 | $3.20 | 256k | not given | 31 | 0.7 s | 100.0% | yes | yes |
| Venice | $0.45 | $3.20 | 256k | fp8 | 35 | 1.1 s | 100.0% | yes | yes |
| featherless-ai | ? | ? | ? | not given | ? | ? | ? | not given | not given |
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": "Qwen/Qwen3.8-27B:near-ai", "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.
- incoai/Qwen3.8-27B-DFlash2535kUnreviewed
- huihui-ai/Huihui-Qwen3.8-27B-abliterated27.8B150kUnreviewed
- Cloudflare/clef27.4B78kUnreviewed
- AEON-7/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF1627.8B74kUnreviewed
- z-lab/Qwen3.8-27B-DFlash242kUnreviewed
- JonathanColetti/Qwen3.8-27B-Uncensored27.4B22kUnreviewed
- prism-ml/Ternary-Bonsai-2-27B-mlx-2bit19kUnreviewed
- JuneoAI/Laplace-1-Atlas27.4B17kUnreviewed
- agentionai/Signal-3.8-27B27.8B14kUnreviewed
- NousResearch/Arete-27B27.4B13kUnreviewed
- Verdugie/Therapy-3.826.9B11kUnreviewed
- orcarouter/Qwen3.8-27B-Uncensored27.8B8.3kUnreviewed
- tencent/Qwen3.8-27B-DSpine-b166.9kUnreviewed
- DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU27.8B6.4kUnreviewed
- mlx-community/Qwen3.8-27B-bf1627.4B6.4kUnreviewed
- ukisai/Swift-1.5-Qwen3.8-27b27.8B5.7kUnreviewed
- RBDash-Team/Xing-VL2.0-Decision26.1B3.9kUnreviewed
- BAAI/AREX-227.4B3.8kUnreviewed
- bottlecapai/ThinkingCap-Qwen3.8-27B27.8B3.6kUnreviewed
- DavidAU/Qwen3.8-27B-Ultimate-Heretic-Uncensored27.8B3.4kUnreviewed
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.