AI Index

Qwen3.6-27B

Language Alibaba Qwen Verified Released Apr 22, 2026
  • In TextImageVideo Out Text
  • Reasoning
  • Reasoning can be off
  • Tool use
  • Structured output
  • Open weights

Prices

3 offerings · prices as listed
Alibaba Cloud Model Studio Official International · USD qwen3.6-27b
Conditions/ 1M tokensInputOutput
Standard $0.60$3.60

Prices from Alibaba Cloud Model Studio pricing (checked Nov 2, 2026)

Alibaba Cloud Model Studio Official China · CNY qwen3.6-27b
Conditions/ 1M tokensInputOutput
Standard ¥3.00¥18.00

Prices from 阿里云百炼：模型价格 (checked Nov 2, 2026)

Standard: the price a request with no conditions pays (the headline price elsewhere on the site).

Details

Built-in tools
Web search
Licence
apache-2.0
Family
qwen
Model ID

Not stated on Alibaba Qwen’s pages: max input, reasoning efforts, max thinking budget, knowledge cutoff, aliases.

Run locally

Weights: Qwen/Qwen3.6-27B
  • Parameters 27.8B
  • Dense
  • Native precision BF16
  • Layers 64
  • KV cache per token 64 KB
  • Native context 256K (model card)
  • Config positions 256K (config.json max_position_embeddings)
  • API context 256K (maker’s API)
My hardware GeForce RTX 4090 24 GB · 1008 GB/s · 64 GB DDR5 dual channel · 89.6 GB/s
On your hardware · GeForce RTX 4090
Runs well Q5_0 (automatic) 8K context 26–38 tok/sOne GPU
WeightsKV cacheOverheadUsable
Needs
≈ 19.2 GB (weights 17.8 + KV cache 0.5 + overhead 0.9)
Usable
21.6 GB (90% of its memory)
Generation speed
26–38 tok/s
Longest context
46K (longer doesn’t fit)
First token
about 0.90–1.35 s for a 2K prompt
PrecisionWeightsNeeds at 8KOn your hardwareLongest context
BF16 (native)maker’s file51.7 GB≈ 54.8Doesn’t fit—
Q8_027.5 GB≈ 29.4Needs system memory 4–6 tok/s—
FP8maker’s file28.7 GB≈ 30.7Needs system memory 4–6 tok/s—
Q6_K21.2 GB≈ 22.8Needs system memory 13–20 tok/s—
Q5_K_M18.4 GB≈ 19.8Runs well 25–37 tok/s36K
Q5_017.8 GB≈ 19.2Runs well 26–38 tok/s46K
Q4_K_M15.7 GB≈ 17.0Runs well 29–43 tok/s81K
Q4_014.6 GB≈ 15.8Runs well 31–47 tok/s101K
IQ4_XS13.7 GB≈ 14.9Runs well 33–49 tok/s115K
Q3_K_Mquality drops12.6 GB≈ 13.8Runs well 36–54 tok/s133K
IQ3_XXSquality drops9.9 GB≈ 10.9Runs well 45–68 tok/s179K
Q2_Kquality drops9.7 GB≈ 10.7Runs well 46–69 tok/s182K
IQ2_XXSquality drops6.7 GB≈ 7.7Runs well 66–98 tok/s231K

Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.

Compared with other hardware

Each at the highest precision it runs. Click one for details.

Smallest that runs it (Q4_K_M, 8K): GPU GeForce RTX 4090; unified memory Apple M6 (24GB or 32GB) · 32 GB.

Run commands

The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.

Memory to fine-tune
Full≈ 418.0 GBDoesn’t fit your hardware
LoRA≈ 56.3 GBDoesn’t fit your hardware
QLoRA≈ 19.1 GBFits your hardware

Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.

Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated

Cited scores

From third-party leaderboards, not measured by us
  • Epoch Capabilities Index 146.7

    #72 of 118 listed models · CI 144.5–148.9

  • Epoch AI · FrontierMath Tiers 1–3 35.1%

    #61 of 84 listed models · ±2.83

    Also listed: reasoning off 34%

  • Epoch AI · GPQA Diamond 85.9%

    #64 of 168 listed models · ±2.48

    Also listed: reasoning off 84.8%

  • Epoch AI · Mystery Game Puzzles 7%

    #72 of 77 listed models · reasoning off · ±2.56

  • Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks'. Licensed CC BY 4.0. · License · as of Oct 28 – Nov 5, 2026
  • EQ-Bench by Samuel J. Paech (eqbench.com; EQ-bench/EQ-bench-site on GitHub), MIT License. · License · as of Oct 24, 2026
  • LiveBench (White et al.), release 2026-06-25, livebench.ai, licensed CC BY-SA 4.0. This board, adapted from it, is shared under CC BY-SA 4.0. · License · as of Jun 25, 2026
  • changes we made

Provider-reported scores

Published by Alibaba Qwen, under the conditions it states
BenchmarkSettingScore
SWE-bench Verifiedinternal agent scaffold, 200K context77.2%
Terminal-Bench 2.0Terminus-2 harness, avg of 5 runs59.3%
MMLU-Prothinking mode86.2%
GPQA Diamondthinking mode87.8%
Humanity's Last Examno tools, thinking mode24%
LiveCodeBenchv6, thinking mode83.9%
AIME 2026full AIME 2026 (I & II), thinking mode94.1%
MMMUthinking mode82.9%
SWE-bench Multilingualinternal agent scaffold, 200K context71.3%
SWE-bench Prorefined benchmark, internal agent scaffold (bash + file-edit), temperature=1.0, top_p=0.95, 200K context53.5%
MMMU-Pro—75.8%

Changes

All changes →
  1. 2 prices dropped on Alibaba Cloud Model Studio, International Output (Reasoning off) $3.60 · Output (Reasoning on) $3.60

Sources

Alibaba Qwen’s official policies: data use, privacy, terms →

Alibaba Qwen · China · qwen.ai

Snapshots, the same family and other models from this provider