Qwen3.6 27B

A third-party technical deep dive published on kie.ai on July 14, 2026 documents what independent testers have measured on Qwen 3.6 27B roughly three weeks after the weights landed on Hugging Face. The model is described in the supplied excerpt as a 27-billion-parameter dense transformer with a hybrid linear-attention The deep dive reports concrete developer-relevant numbers: NVFP4 quantization hits MMLU accuracy of 0.8446 and delivers roughly 2.6–2.86× decode speedup over BF16 in a vLLM benchmark. On DGX Spark hardware, community testers measured 28–33 tokens/second single-session throughput on the NVFP4 build via vLLM 0.24.0, whil

A DGX Spark community benchmark compared Qwen/Qwen3.6-27B in FP16 against nvidia/Qwen3.6-27B-NVFP4 using vLLM 0.24.0 and lm-eval on MMLU. The conclusion reported by the poster is that NVFP4 quantization reaches FP16-level accuracy on the model. The thread confirms Qwen3.6-27B runs on vLLM with an OpenAI-compatible API and that NVIDIA has published a NVFP4 weight variant alongside the original Hugging Face release. This gives developers a concrete quantization option that preserves accuracy at lower precision on Blackwell-class hardware.

Qwen (Tongyi Lab), an AI research team at the Chinese AI company Alibaba, has released ' Qwen3.6-27B ,' a multimodal AI model with 27 billion parameters. It is licensed under the open Apache License 2.0 and can be used commercially. Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model https://qwen.ai/blog?id=qwen3.6

Alibaba's new open-source model Qwen3.6-27B beats its 15-times-larger predecessor across coding benchmarks with just 27 billion parameters.

Alibaba's Qwen3.6-27B crushes coding benchmarks, fueling coder variant buzz

Alibaba Qwen Team Releases Qwen3.6-27B: A Dense Open-Weight Model Outperforming 397B MoE on Agentic Coding Benchmarks

Alibaba's Qwen team released Qwen3.6-27B as an open-source dense 27-billion-parameter multimodal model on April 21, 2026, following Qwen3.6-Plus and Qwen3.6-35B-A3B. It supports both thinking and non-thinking modes and is available on Qwen Studio, Hugging Face, and ModelScope, with API access coming via Alibaba Cloud Model Studio. On coding benchmarks, Qwen3.6-27B posts 77.2 on SWE-bench Verified, 53.5 on SWE-bench Pro, 59.3 on Terminal-Bench 2.0, and 48.2 on SkillsBench, surpassing the Qwen3.5-397B-A17B MoE baseline on each test. It also reaches 87.8 on GPQA Diamond and is designed for straightforward dense deployment without MoE routing complexity.

Released on April 22, 2026 under Apache 2.0, Qwen3.6-27B is a dense 27B-parameter model that reportedly surpasses the much larger Qwen3.5-397B-A17B on agentic coding benchmarks, including 77.2 on SWE-bench Verified. The result challenges the assumption that flagship coding capability requires massive parameter counts, highlighting architectural efficiency and training-data quality over raw scale. For developers this means near-flagship coding performance in an open-weights package. Technically, it uses a 64-layer hybrid architecture combining Gated DeltaNet and Gated Attention layers, with a context window of 262,144 tokens extensible to roughly 1,010,000 tokens. The model supports multimodal thinking and non-thinking modes within a single checkpoint and ships in BF16 and fine-grained FP8 (block size 128) weight variants on Hugging Face. Native vision-language support for image and video input is included alongside text, broadening its applicability for multimodal applications.

Qwen3.6 27B is a dense 27-billion-parameter language model attributed to the Qwen Team at Alibaba, with an OpenRouter catalog page listing an April 27, 2026 release. The model is described as supporting hybrid multimodal inputs spanning text, image, and video, along with a 262,144-token context window. It is released under the Apache 2.0 license and supports 201 languages and dialects, positioning it as an open-weights release aimed at broad deployment. The catalog page highlights a built-in thinking mode for extended reasoning, with thinking context preserved across conversation history. It targets agentic coding and reasoning workloads, citing particular strength in repository-level code comprehension, front-end development workflows, and multi-step problem solving. OpenRouter lists the model at $0.30 per million input tokens and $2 per million output tokens as reference pricing, though the page primarily aggregates per-provider throughput and routing data.

Alibaba released Qwen3.6-27B on April 22, 2026 as the first open-weights variant in the Qwen3.6 family and the latest entry in the Qwen Coder line. It is a dense 27-billion-parameter coding-focused model distributed under the Apache 2.0 license, making state-of-the-art coding capabilities freely available to developers. The release landed six days after Qwen 3.6-35B-A3B and continues Alibaba's roughly 78-day Qwen Coder cadence. The model offers a 262,144-token native context window that can be extended up to roughly one million tokens for long-context workloads. This positions Qwen3.6-27B as a versatile open-weights option for tasks that require substantial document or codebase ingestion. The lineage information is drawn from the candidate's release-tracker page, which cites Hugging Face and the Qwen GitHub as its underlying sources.

A 67AI Lab deep dive documents Qwen3.6-27B's hybrid architecture, organized as 64 layers in 16 macro-blocks of three Gated DeltaNet plus FFN layers followed by one Gated Attention plus FFN layer. The model carries a 5120 hidden dimension, 248,320-token embedding, 262,144-token native context extensible to about 1,010,000, and multi-token prediction training. The post frames the design as selective full attention layered with cheaper linear-attention-style computation to keep KV-cache costs manageable while supporting agentic coding and long-context reasoning. It positions mid-size dense models as viable against much larger MoE systems when compute allocation and post-training are tuned well.