Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Models › Qwen3.6 family

Qwen3.6-27B

Hugging Face ↗ Follow changes

Qwen Apache 2.0 out 6 months ago

27.8B parameters, images and text in, text out, 256K tokens of context. 9 providers serve it from $0.20 per million input tokens.

Verified 2026-10-25 Rank #32 in LLMs · score 112.9 method v1 Sources 4 Downloads 30d 2.4M · #57 of 619 tracked

Changed this week: 2 new quantized builds on the Hub (749 in all).2026-10-24 · 1 new quantized build on the Hub (748 in all).2026-10-24

Can I run it?17GB at INT4Estimatefits a 24 GB cardmethodCommunity: Club-3090 Recipes for serving QWEN3.6 27B locally on RTX 3090s →
What does it cost?$0.20/1M incheapest of 6Reka2026-10-24
Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24
Do people use it?2.4Mdownloads, 30 days#57 of 619 we track · top 9%+5 likes, downloads +5% this week2,335 likesHugging Face2026-10-24

Worth reading

Picked by rule from 36 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Qwen3.6-27B supports the preserve_thinking feature, which is recommended for agentic tasks to retain reasoning content.

From Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model · qwen.ai · 2026-04-22
The lab's own post

Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model

Qwen3.6-27B is a 27-billion-parameter dense multimodal model released on 2026/04/21 that delivers flagship-level agentic coding performance by surpassing the previous-generation open-source flagship Qwen3.5-397B-A17B across all major coding benchmarks.

  • The model achieves a 53.5 score on SWE-bench Pro, significantly outperforming the larger Qwen3.5-397B-A17B which scores 50.9.
  • Qwen3.6-27B supports the preserve_thinking feature, which is recommended for agentic tasks to retain reasoning content.
qwen.ai 2026-04-22 993 HN pts · 458 comments 2.2k words

25 best of 36 tracked

The 4 picks above, plus these 21. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

  1. ArticleQwen3.6-35B-A3B: Agentic coding power, now open to allqwen.ai · 2026-04-16 · official
  2. ArticleQwen3.6-Max-Preview: Smarter, Sharper, Still Evolvingqwen.ai · 2026-04-20 · official
  3. ArticleQwen3.6-Plus: Towards real world agentsqwen.ai · 2026-04-02 · official
  4. ArticleQwen3.6-35B-A3B on my laptop drew me a better pelican than Claude Opus 4.7simonwillison.net · 2026-04-16
  5. CodeShow HN: Qwen3.6-35B-A3B on a 16 GB M1 Pro with SSD-streamed MoEgithub.com · 2026-07-17
  6. CodeShow HN: Samosa Chat - Run Qwen3.6-35B-A3B Locally on a 16 GB Macgithub.com · 2026-07-15
  7. CodeGufo-Qwen3.6-35B-A3B-Q6dense – 3095tok/s prefill; 190 tok/s decode on Strix Halogithub.com · 2026-10-03
  8. CodeIris-mini – open search agent on Qwen3.6-35B-A3Bgithub.com · 2026-09-13
  9. ArticleQwen3.6-27B: Flagship-Level Coding in a 27B Dense Modelsimonwillison.net · 2026-04-27 · 5 HN pts
  10. CodeNInfer: 271 tok/s (543 with speculative) for Qwen3.6-35B-A3B on one RTX 5090github.com · 2026-07-23
  11. CodeQwen3.6-35B-A3B speculative decoding is net-negative on RTX 3090github.com · 2026-04-21
  12. CodeShow HN: Modernizing my old PhD work in an evening with little Qwen3.6 MoEgithub.com · 2026-05-21 · 3 HN pts
  13. CodeGB10 Solution Atlas is now open source, <2min cold start 100 tok/s Qwen3.6-FP8github.com · 2026-05-06 · 2 HN pts
  14. CodeMessy Model Bench Tests; Qwen3.6-27B vs. Coder-Nextgithub.com · 2026-05-03 · 4 HN pts
  15. ArticleLocal LLMs on a Ryzen 8700G iGPU: 13-15 tok/s on gemma4, 9-12 on qwen3.6loomcycle.dev · 2026-06-30 · 4 HN pts
  16. ArticleQuantization hurts knowledge nonlinearly – Qwen3.6 27B case studyquesma.com · 2026-08-03 · 2 HN pts
  17. DiscussionNorm-preserving abliteration on Qwen3.6-35B-A3B with 0% refusalreddit.com · 2026-07-03
  18. CodeShow HN: Qwen Lens Studio – multimodal app on Qwen3.6-35B-A3B, runs on Ollamagithub.com · 2026-04-22
  19. CodeLocally Uncensored v2.3.3 – local AI with Qwen3.6 and remote accessgithub.com · 2026-04-16
  20. DiscussionShow HN: Open Access Qwen3.6-35B-A3B-UD-Q5_K_M with TurboQuantnews.ycombinator.com · 2026-04-17
  21. CodeClub-3090 Recipes for serving QWEN3.6 27B locally on RTX 3090sgithub.com · 2026-05-05

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

Prices for the same weights vary 2×

6 providers with a live price, from $0.20 to $0.47 per million input tokens; the median is $0.32. One dot per provider, cheapest in teal.

$0.010$0.10$1.00$10.00Reka: $0.20 in / $2.00 out per 1M · observed 2026-10-24 · https://openrouter.ai/qwen/qwen3.6-27bSiliconFlow: $0.30 in / $3.20 out per 1M · observed 2026-10-24 · https://openrouter.ai/qwen/qwen3.6-27bDeepInfra: $0.32 in / $3.20 out per 1M · observed 2026-10-25 · https://huggingface.co/api/models/Qwen/Qwen3.6-27B?expand[]=inferenceProviderMappingVenice: $0.33 in / $3.25 out per 1M · observed 2026-10-24 · https://openrouter.ai/qwen/qwen3.6-27bAlibaba: $0.45 in / $2.70 out per 1M · observed 2026-10-24 · https://openrouter.ai/qwen/qwen3.6-27bOVHcloud: $0.47 in / $3.19 out per 1M · observed 2026-10-25 · https://huggingface.co/api/models/Qwen/Qwen3.6-27B?expand[]=inferenceProviderMappingcheapest $0.20median $0.32
$0.010$0.10$1.00$10.00Reka: $0.20 in / $2.00 out per 1M · observed 2026-10-24 · https://openrouter.ai/qwen/qwen3.6-27bSiliconFlow: $0.30 in / $3.20 out per 1M · observed 2026-10-24 · https://openrouter.ai/qwen/qwen3.6-27bDeepInfra: $0.32 in / $3.20 out per 1M · observed 2026-10-25 · https://huggingface.co/api/models/Qwen/Qwen3.6-27B?expand[]=inferenceProviderMappingVenice: $0.33 in / $3.25 out per 1M · observed 2026-10-24 · https://openrouter.ai/qwen/qwen3.6-27bAlibaba: $0.45 in / $2.70 out per 1M · observed 2026-10-24 · https://openrouter.ai/qwen/qwen3.6-27bOVHcloud: $0.47 in / $3.19 out per 1M · observed 2026-10-25 · https://huggingface.co/api/models/Qwen/Qwen3.6-27B?expand[]=inferenceProviderMappingcheapest $0.20 · median $0.32 (white tick)
#Cheapest 5In /1MOut /1MContextQuantSeen
1 Reka $0.20 $2.00 256K unknown Oct 24
2 SiliconFlow $0.30 $3.20 256K fp8 Oct 24
3 DeepInfra $0.32 $3.20 256K fp8 Oct 25
4 Venice $0.33 $3.25 250K fp8 Oct 24
5 Alibaba $0.45 $2.70 256K unknown Oct 24

All 9 providers →

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

Widely served (9 API providers); heavily run locally (749 quantized builds); a big base for builders (1,003 fine-tunes, adapters and merges).

  • As a service9Alibaba, Chutes, DeepInfra, Featherless +5
  • On own hardware749quantized builds
  • Built on1,003415 fine-tunes · 563 adapters · 25 merges
  • In apps100+Hugging Face Spaces

Top Spaces running it: embedl/hfviewer ♥ 181 · dromerosm/groq-chatbot ♥ 32 · lablab-ai-amd-developer-hackathon/OncoAgent ♥ 10 · build-small-hackathon/ouroboros-kernel-mint ♥ 8

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 64 GB Mac (unified memory) at INT4

About 17 GB at INT4, 33 GB at FP8, 67 GB at BF16: weights only, so leave room for the KV cache.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 17 GB of weights (estimate; KV cache and activations extra)INT417 GBFP8: about 33 GB of weights (estimate; KV cache and activations extra)FP833 GBBF16: about 67 GB of weights (estimate; KV cache and activations extra)BF1667 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 17 GB of weights (estimate; KV cache and activations extra)INT417 GBFP8: about 33 GB of weights (estimate; KV cache and activations extra)FP833 GBBF16: about 67 GB of weights (estimate; KV cache and activations extra)BF1667 GB

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

10 dated articles and 13 repo commits so far

Teal is the weights (2026-04-21), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.

MayJunJulAugSepOct2026-04-24: 1 repo commit2026-04-22: 11 repo commits2026-04-21: 1 repo commitQwen3.6-27B: Flagship-Level Coding in a 27B Dense Model · 2026-04-22Running Claude Code Offline on an M3 Pro with Qwen3.6 · 2026-06-11Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model · 2026-04-27ThinkingCap – we fine-tuned Qwen3.6-27B to reduce unnecessary reasoning · 2026-08-25Show HN: Modernizing my old PhD work in an evening with little Qwen3.6 MoE · 2026-05-21GB10 Solution Atlas is now open source, <2min cold start 100 tok/s Qwen3.6-FP8 · 2026-05-06Messy Model Bench Tests; Qwen3.6-27B vs. Coder-Next · 2026-05-03Show HN: Qwen3.6 27B on a DGX Spark writing code that ships · 2026-07-07Local LLMs on a Ryzen 8700G iGPU: 13-15 tok/s on gemma4, 9-12 on qwen3.6 · 2026-06-30Quantization hurts knowledge nonlinearly – Qwen3.6 27B case study · 2026-08-03Weights 2026-04-21 · 2026-04-21Weights 2026-04-21now · 9 providers
MayJunJulAugSepOct2026-04-24: 1 repo commit2026-04-22: 11 repo commits2026-04-21: 1 repo commitQwen3.6-27B: Flagship-Level Coding in a 27B Dense Model · 2026-04-22Running Claude Code Offline on an M3 Pro with Qwen3.6 · 2026-06-11Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model · 2026-04-27ThinkingCap – we fine-tuned Qwen3.6-27B to reduce unnecessary reasoning · 2026-08-25Show HN: Modernizing my old PhD work in an evening with little Qwen3.6 MoE · 2026-05-21GB10 Solution Atlas is now open source, <2min cold start 100 tok/s Qwen3.6-FP8 · 2026-05-06Messy Model Bench Tests; Qwen3.6-27B vs. Coder-Next · 2026-05-03Show HN: Qwen3.6 27B on a DGX Spark writing code that ships · 2026-07-07Local LLMs on a Ryzen 8700G iGPU: 13-15 tok/s on gemma4, 9-12 on qwen3.6 · 2026-06-30Quantization hurts knowledge nonlinearly – Qwen3.6 27B case study · 2026-08-03Weights 2026-04-21 · 2026-04-21Weights 2026-04-21now · 9 providers

What people built on it

1,752 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

1,752 derivatives on the Hub, led by 749 quantized

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

749 quantized (Hugging Face model tree)415 fine-tunes (Hugging Face model tree)25 merges (Hugging Face model tree)563 adapters (Hugging Face model tree)749 quantized415 fine-tunes25 merges563 adapters
749 quantized (Hugging Face model tree)415 fine-tunes (Hugging Face model tree)25 merges (Hugging Face model tree)563 adapters (Hugging Face model tree)749 quantized415 fine-tunes25 merges563 adapters
GGUF unsloth/Qwen3.6-27B-MTP-GGUF for llama.cpp, Ollama, LM Studio 7M downloads MLX lmstudio-community/Qwen3.6-27B-MLX-8bit for Apple silicon 3.4M downloads FP8 Qwen/Qwen3.6-27B-FP8 for vLLM, SGLang 36.3M downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

7 rows

Benchmarks 7 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (7)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.
BenchmarkValueSetting / notesReported byDate
author-reported NL2Repo · Write code with agents36.2Qwen3.6-27Bmodel author2026-10-07
author-reported QwenClawBench53.4Qwen3.6-27Bmodel author2026-10-07
author-reported QwenWebBench1487Qwen3.6-27Bmodel author2026-10-07
author-reported SWE-bench Multilingual · Write code with agents71.3Qwen3.6-27Bmodel author2026-10-07
author-reported SWE-bench Pro · Write code with agents53.5Qwen3.6-27Bmodel author2026-10-07
author-reported SWE-bench Verified · Write code with agents77.2Qwen3.6-27Bmodel author2026-10-07
author-reported Terminal-Bench · Write code with agents59.3Qwen3.6-27Bmodel author2026-10-07

What the card says

2 cited sentences

Summary generated, cited

Qwen3.6-27B is an image-text-to-text causal language model with a vision encoder containing 27.8B parameters.[1][2] The model features a native context length of 262,144 tokens and is released under the Apache 2.0 license.[3][4]

  1. parameters human
  2. “Type: Causal Language Model with Vision Encoder”
  3. “Context Length: 262,144 natively”
  4. license

Generated 2026-10-07 14:59 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

6 use cases

What it is good at 6 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

20 fields

Listed in LLMs

Facts

Pipeline tagimage-text-to-text
Librarytransformers
ArchitecturesQwen3_5ForConditionalGeneration
Model typeqwen3_5
Dtypebfloat16
Params by dtypeBF16: 27,781,427,952
Size bandL
MoEno
Context length262,144 (config.json max_position_embeddings)
LicenseApache 2.0 (apache-2.0)
Hub snapshot2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · Qwen3_5ForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3_5ForConditionalGeneration · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for Qwen3.6-27B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)67 GB · Single high-end GPU (<80 GB)
FP8 (1 byte/param)33 GB · Single high-end GPU (<80 GB)
INT4 (0.5 bytes/param)17 GB · Single GPU (<24 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

40 variants

Lineage

Original model Family: Qwen/Qwen3.6 1,752 derivatives on the Hub

quantized: 749finetune: 415adapter: 563merge: 25

Top variants (40 shown, ranked by downloads)

Relation
Quant
Size
40 rows
Variant Relation Quant Params Band License Author Likes Downloads
Qwen/Qwen3.6-27B-FP8 quantized -FP8 27.8B L apache-2.0 Qwen 375 36.3M
unsloth/Qwen3.6-27B-NVFP4 quantized NVFP4 21.2B L apache-2.0 unsloth 286 10.1M
cyankiwi/Qwen3.6-27B-AWQ-INT4 quantized AWQ 27.8B L apache-2.0 cyankiwi 113 9.4M
unsloth/Qwen3.6-27B-MTP-GGUF quantized GGUF — — apache-2.0 unsloth 1.3k 7M
unsloth/Qwen3.6-27B-GGUF quantized GGUF — — apache-2.0 unsloth 971 6.4M
Lorbus/Qwen3.6-27B-int4-AutoRound quantized int4 6.28B M apache-2.0 Lorbus 129 5.4M
QuantTrio/Qwen3.6-27B-AWQ quantized AWQ 27.8B L apache-2.0 QuantTrio 28 4.2M
prism-ml/Ternary-Bonsai-27B-mlx-2bit finetune mlx 27.4B L apache-2.0 prism-ml 216 4.1M
prism-ml/Bonsai-27B-mlx-1bit finetune mlx 1.72B S apache-2.0 prism-ml 281 4.1M
nvidia/Qwen3.6-27B-NVFP4 quantized NVFP4 18.2B L apache-2.0 nvidia 446 4.1M
prism-ml/Bonsai-27B-gguf quantized gguf — — apache-2.0 prism-ml 928 3.8M
lmstudio-community/Qwen3.6-27B-MLX-8bit quantized MLX 27.4B L apache-2.0 lmstudio-community 17 3.4M
lmstudio-community/Qwen3.6-27B-MLX-4bit quantized MLX 27.4B L apache-2.0 lmstudio-community 17 3.3M
sakamakismile/Qwen3.6-27B-Text-NVFP4-MTP quantized NVFP4 16.7B L apache-2.0 sakamakismile 82 3.3M
lmstudio-community/Qwen3.6-27B-MLX-6bit quantized MLX 27.4B L apache-2.0 lmstudio-community 3 3.1M
lmstudio-community/Qwen3.6-27B-MLX-5bit quantized MLX 27.4B L apache-2.0 lmstudio-community 0 3.1M
prism-ml/Ternary-Bonsai-27B-gguf quantized gguf — — apache-2.0 prism-ml 1.4k 2.4M
HauhauCS/Qwen3.6-27B-Uncensored-HauhauCS-Aggressive quantized GGUF — — apache-2.0 HauhauCS 740 2.1M
lmstudio-community/Qwen3.6-27B-GGUF quantized GGUF — — apache-2.0 lmstudio-community 32 2M
groxaxo/Qwen3.6-27B-GPTQ-Pro-4bit quantized GPTQ 27.4B L apache-2.0 groxaxo 39 1.8M
cyankiwi/Qwen3.6-27B-AWQ-BF16-INT4 quantized AWQ 27.8B L apache-2.0 cyankiwi 42 1.5M
shawnw3i/Huihui-Qwen3.6-27B-abliterated-AWQ-MTP quantized AWQ 6.28B M apache-2.0 shawnw3i 15 1.5M
Intel/Qwen3.6-27B-int4-AutoRound quantized int4 6.28B M apache-2.0 Intel 46 1.1M
Jackrong/Qwopus3.6-27B-v1-preview-GGUF quantized GGUF — — apache-2.0 Jackrong 123 1.1M
Jackrong/Qwopus3.6-27B-v2-MTP-GGUF adapter GGUF — — apache-2.0 Jackrong 392 1M
michaelw9999/Qwen3.6-27B-NVFP4-MTP-GGUF quantized NVFP4 — — — michaelw9999 63 919.4k
rdtand/Qwen3.6-27B-PrismaSCOUT-Blackwell-NVFP4-BF16-vllm quantized NVFP4 15.5B L apache-2.0 rdtand 44 866.8k
batiai/Qwen3.6-27B-GGUF quantized GGUF — — apache-2.0 batiai 8 849.4k
HauhauCS/Qwen3.6-27B-Uncensored-HauhauCS-Balanced quantized GGUF — — apache-2.0 HauhauCS 218 813.2k
KyleHessling1/Qwopus3.6-27B-Fusion-GGUF quantized GGUF — — other KyleHessling1 78 771.4k
bytkim/Qwen3.6-27B-MTP-pi-tune-GGUF finetune GGUF — — apache-2.0 bytkim 126 717.7k
bartowski/Qwen_Qwen3.6-27B-GGUF quantized GGUF — — apache-2.0 bartowski 89 576.9k
Alennnndy/Qwen3.6-27B-AWQ-4bit quantized AWQ 27.7B L apache-2.0 Alennnndy 7 398.3k
huihui-ai/Huihui-Qwen3.6-27B-abliterated-MTP-GGUF quantized GGUF — — apache-2.0 huihui-ai 99 354k
btbtyler09/Qwen3.6-27B-GPTQ-4bit quantized GPTQ 28.2B L apache-2.0 btbtyler09 2 283.9k
mlx-community/Qwen3.6-27B-4bit quantized -4bit 27.4B L apache-2.0 mlx-community 52 281k
unsloth/Qwen3.6-27B finetune — 27.8B L apache-2.0 unsloth 45 262.7k
QuantTrio/Qwen3.6-27B-AWQ-6Bit quantized AWQ 27.8B L apache-2.0 QuantTrio 17 261k
lhca521/Huihui-Qwen3.6-27B-abliterated-AWQ quantized AWQ 6.28B M apache-2.0 lhca521 1 245.5k
ggml-org/Qwen3.6-27B-GGUF quantized GGUF — — apache-2.0 ggml-org 27 224.4k

Repository changelog

13 commits
changed in the last 7 daysOther: 2026-10-24, 2026-10-24all changes

Repository changelog 13 commits

DateCommitSHA
2026-04-24 fix: rename response variable to chat_response in README code example 6a9e13b
2026-04-22 Update README.md 5d316fa
2026-04-22 Update README.md b4e2a56
2026-04-22 Update README.md 5f68465
2026-04-22 Update README.md 3ba3f15
2026-04-22 Upload LICENSE e61d73f
2026-04-22 Update tokenizer_config.json a1bdd8d
2026-04-22 Update generation_config.json b7897d5
2026-04-22 Update README.md 3322bcd
2026-04-22 Delete .mv ea24fab
2026-04-22 Delete .msc d45cab9
2026-04-22 Upload folder using huggingface_hub 1b559cf
2026-04-21 initial commit 70215ea

All providers

9 providers · 9 endpoints

Where to run 9 providers · 9 endpoints

Hosted endpoints observed for Qwen3.6-27B, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.20/M.

ProviderModel refContextQuant$/M in$/M outtok/sTTFTToolsStatusObserved
Reka
openrouter
 qwen/qwen3.6-27b 256K unknown $0.20 $2.00 — — yes live 2026-10-24
Chutes
openrouter
 qwen/qwen3.6-27b 256K fp8 $0.25 $2.00 — — yes status -2 2026-10-24
SiliconFlow
openrouter
 qwen/qwen3.6-27b 256K fp8 $0.30 $3.20 — — yes live 2026-10-24
DeepInfra
hf, deepinfra, openrouter
 Qwen/Qwen3.6-27B 256K fp8 $0.32 $3.20 57.1 250 ms yes live 2026-10-25
Phala
openrouter
 qwen/qwen3.6-27b 256K unknown $0.32 $3.25 — — yes status -2 2026-10-24
Venice
openrouter
 qwen/qwen3.6-27b 250K fp8 $0.33 $3.25 — — yes live 2026-10-24
Alibaba
openrouter
 qwen/qwen3.6-27b 256K unknown $0.45 $2.70 — — yes live 2026-10-24
OVHcloud
hf
 Qwen3.6-27B 256K — $0.47 $3.19 54.5 542 ms yes live 2026-10-25
Featherless
hf
 Qwen/Qwen3.6-27B — — — — — — — live 2026-10-25

Articles and coverage

36 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official1

Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model
qwen.ai2026-04-22officialauthorityHN front pageHN 993 pts · 458 comments2,168 words
Qwen3.6-27B is a 27-billion-parameter dense multimodal model released on 2026/04/21 that delivers flagship-level agentic coding performance by surpassing the previous-generation open-source flagship Qwen3.5-397B-A17B across all major coding benchmarks. It achieves a SWE-bench Verified score of 77.2 compared to 76.2 for the larger model and is available as open weights on Hugging Face and ModelScope for self-hosting. The model supports both vision-language thinking and non-thinking modes in a single unified checkpoint.written, reviewed, cited
  • The model achieves a 53.5 score on SWE-bench Pro, significantly outperforming the larger Qwen3.5-397B-A17B which scores 50.9.
  • Qwen3.6-27B supports the preserve_thinking feature, which is recommended for agentic tasks to retain reasoning content.

Analysis9

Running Claude Code Offline on an M3 Pro with Qwen3.6
har-ki.github.io2026-06-11about the Qwen3.6 familyHN 18 pts · 10 comments2,250 words
Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model
simonwillison.net2026-04-27authorityHN 5 pts · 1 comments
ThinkingCap – we fine-tuned Qwen3.6-27B to reduce unnecessary reasoning
bottlecapai.com2026-08-25HN 3 pts · 0 comments
Show HN: Modernizing my old PhD work in an evening with little Qwen3.6 MoE
github.com2026-05-21authorityabout the Qwen3.6 familyHN 3 pts · 0 comments
GB10 Solution Atlas is now open source, <2min cold start 100 tok/s Qwen3.6-FP8
github.com2026-05-06authorityabout the Qwen3.6 familyHN 2 pts · 1 comments
Messy Model Bench Tests; Qwen3.6-27B vs. Coder-Next
github.com2026-05-03authorityHN 4 pts · 0 comments
Show HN: Qwen3.6 27B on a DGX Spark writing code that ships
devashish.me2026-07-07HN 2 pts · 1 comments
Local LLMs on a Ryzen 8700G iGPU: 13-15 tok/s on gemma4, 9-12 on qwen3.6
loomcycle.dev2026-06-30about the Qwen3.6 familyHN 4 pts · 0 comments
Quantization hurts knowledge nonlinearly – Qwen3.6 27B case study
quesma.com2026-08-03HN 2 pts · 0 comments
All coverage tracked for this model (36) — append-only, nothing is ever removed
TitleSourcePublishedFirst seenSignals
Qwen 3.6 27B · LLM Model Review | CrucibleMarkcruciblemark.com—2026-10-07exa
Gufo-Qwen3.6-35B-A3B-Q6dense – 3095tok/s prefill; 190 tok/s decode on Strix Halogithub.com2026-10-032026-10-07hn
Iris-mini – open search agent on Qwen3.6-35B-A3Bgithub.com2026-09-132026-10-07hn
ThinkingCap – we fine-tuned Qwen3.6-27B to reduce unnecessary reasoningbottlecapai.com2026-08-252026-10-07hn
Quantization hurts knowledge nonlinearly – Qwen3.6 27B case studyquesma.com2026-08-032026-10-07hn
NInfer: 271 tok/s (543 with speculative) for Qwen3.6-35B-A3B on one RTX 5090github.com2026-07-232026-10-07hn
Show HN: Qwen3.6-35B-A3B on a 16 GB M1 Pro with SSD-streamed MoEgithub.com2026-07-172026-10-07hn
Show HN: Samosa Chat - Run Qwen3.6-35B-A3B Locally on a 16 GB Macgithub.com2026-07-152026-10-07hn
Qwen 3.6 27B Benchmark Deep Dive: The Measured Numberskie.ai2026-07-152026-10-07exa
Show HN: Qwen3.6 27B on a DGX Spark writing code that shipsdevashish.me2026-07-072026-10-07hn
Norm-preserving abliteration on Qwen3.6-35B-A3B with 0% refusalreddit.com2026-07-032026-10-07hn
Local LLMs on a Ryzen 8700G iGPU: 13-15 tok/s on gemma4, 9-12 on qwen3.6loomcycle.dev2026-06-302026-10-07hn
Running Claude Code Offline on an M3 Pro with Qwen3.6har-ki.github.io2026-06-112026-10-07hn
Qwen3.6, reviewed — benchrbenchr.org2026-05-302026-10-07exa
A Comprehensive Analysis of the Open-Weight Frontier (May 2026)deepresearch.ninja2026-05-282026-10-07hn
Show HN: Modernizing my old PhD work in an evening with little Qwen3.6 MoEgithub.com2026-05-212026-10-07hn
GB10 Solution Atlas is now open source, <2min cold start 100 tok/s Qwen3.6-FP8github.com2026-05-062026-10-07hn
Club-3090 Recipes for serving QWEN3.6 27B locally on RTX 3090sgithub.com2026-05-052026-10-07hn
Messy Model Bench Tests; Qwen3.6-27B vs. Coder-Nextgithub.com2026-05-032026-10-07hn
Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Modelsimonwillison.net2026-04-272026-10-07hn
Qwen3.6-27B Deep Dive: Why This Mid-Size Dense Model Works So Well — 67AI Lab67ailab.com2026-04-232026-10-07exa
Qwen 3.6-27B Deep Dive: Alibaba's Dense Flagship with Hybrid Thinking | LocalClawlocalclaw.io2026-04-232026-10-07exa
Qwen3.6-27B Explained: Agentic Coding, Hybrid Architecture, Benchmarks & API on Qubrid AI - Qubrid AIqubrid.com2026-04-232026-10-07exa
Show HN: Qwen Lens Studio – multimodal app on Qwen3.6-35B-A3B, runs on Ollamagithub.com2026-04-222026-10-07hn
Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Modelqwen.ai official2026-04-222026-10-07hn
Qwen3.6-27Btwitter.com2026-04-222026-10-07hn
Qwen3.6-27B: 27B Coding Model Beats 397B Rival | HokAIhokai.io2026-04-222026-10-07exa
Qwen3.6-35B-A3B speculative decoding is net-negative on RTX 3090github.com2026-04-212026-10-07hn
Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolvingqwen.ai official2026-04-202026-10-07hn
Show HN: Open Access Qwen3.6-35B-A3B-UD-Q5_K_M with TurboQuantnews.ycombinator.com2026-04-172026-10-07hn
Qwen3.6-35B-A3B draws a better pelican than Opus 4.7twitter.com2026-04-162026-10-07hn
Qwen3.6-35B-A3B on my laptop drew me a better pelican than Claude Opus 4.7simonwillison.net2026-04-162026-10-07hn
Locally Uncensored v2.3.3 – local AI with Qwen3.6 and remote accessgithub.com2026-04-162026-10-07hn
Qwen3.6-35B-A3B：Now Open-Sourcetwitter.com2026-04-162026-10-07hn
Qwen3.6-35B-A3B: Agentic coding power, now open to allqwen.ai official2026-04-162026-10-07hn
Qwen3.6-Plus: Towards real world agentsqwen.ai official2026-04-022026-10-07hn

Similar models

3 models

Similar models

  • Qwen3.8-27B · 27.8B
    same size band (L), same task (image-text-to-text), more providers, permissive license
  • Qwen3.5-27B · 27.8B
    same size band (L), same task (image-text-to-text), permissive license
  • gemma-3-27b-it · 27.4B
    same size band (L), same task (image-text-to-text)

Sources

4 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:32 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 14:59 UTC · Last verified 2026-10-25

Follow Qwen3.6-27B

Watch this model

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.