All models › alibaba

Qwen3.5-27B

alibaba · Apache 2.0 license · on Arena since 26 Feb 2026

Lab chat Download
Arena score1409#142 of 416 in Chat, paid models included
Size27.8 Bparameters
4-bit download17 GBweights only, Q4_K_M
Max context256Ktokens, from the model config

Where to use it for free

Arena scores

ArenaScoreRankVotes
Chat1409#142 of 41629K
Coding1358#110 of 1448K
Vision1241#72 of 16530K

Rank among all models, paid ones included. Source: Arena leaderboard dataset.

Speed on popular hardware

HardwareGPU memoryFits?Speed
RTX 3060 12 GB12 GBGPU + 64 GB RAM6.9 tok/s
RTX 4060 Ti 16 GB16 GBGPU + 64 GB RAM8.1 tok/s
RTX 309024 GBFits in GPU34 tok/s
RTX 409024 GBFits in GPU37 tok/s
RTX 509032 GBFits in GPU66 tok/s
RX 7900 XTX24 GBFits in GPU35 tok/s
2 × RTX 309048 GBFits in GPU33 tok/s
4 × RTX 309096 GBFits in GPU32 tok/s
Mac M4 Max 128 GB96 GBFits in GPU20 tok/s
Ryzen AI Max+ 395 128 GB96 GBFits in GPU9.4 tok/s
8 × H100 SXM640 GBFits in GPU102 tok/s

Estimate at 4-bit with 32K context, 64 GB of DDR5 for offload. Speed ≈ memory bandwidth × 70% ÷ data read per token; real numbers vary by about ±30%.

GPU memory needed

Level8K context32K context128K contextQuality vs full
Q3_K_M15 GB17 GB23 GB+7.2% noticeable loss
Q4_K_M18 GB20 GB26 GB+1.9% small loss
Q5_K_M21 GB23 GB29 GB+0.6% very small loss
Q6_K24 GB26 GB32 GB+0.2% near-lossless
Q8_031 GB32 GB39 GB≈ 0% same as full
F1657 GB58 GB65 GB≈ 0% same as full

Weights + context cache (from the model's architecture) + runtime. Quality: increase in perplexity against the full model, from llama.cpp tests, scaled by model size. Simulate your own setup.

Chat score over time

Qwen3.5-27B 1409Best paid model

Arena score from 17 Apr 2026 to 15 Oct 2026.

Vision score over time

Qwen3.5-27B 1241Best paid model

Arena score from 10 Apr 2026 to 9 Oct 2026.

Similar free models in Chat