Qwen 3.6 27B

Within the Qwen 3.6 generation, the 27B variant is described as a dense model that fits into roughly 24 GB of VRAM at Q4 quantization, making it practical for high-end consumer GPUs rather than data-center setups. A community quantization on Hugging Face points to the official Qwen 3.6 27B release published under the Qwen organization, and the upstream weights are distributed under an Apache 2.0 license. Because the dense architecture avoids the multi-expert routing used by larger MoE peers, latency and memory behavior tend to be more predictable, which helps local deployments that rely on single-GPU serving and straightforward batching.

A third-party 2026 comparison of local LLMs reports the 27B Qwen 3.6 checkpoint at 77.2% on SWE-bench, positioning it as a dense coding leader against same-era Llama 4 and Mistral alternatives and as a balanced pick for multilingual development workloads. The same comparison highlights the model's broad language coverage and its ability to deliver competitive reasoning quality without the heavier memory footprint of long-context MoE options, suggesting a practical fit for developers who want a single local model that handles code generation, instruction following, and general reasoning on a 24 GB card.

Venice AIqwen3-6-27bqwen

Open provider page Provider docs

Quick Info

Provider: Venice AI
Model key: qwen3-6-27b
Release date: Apr 24, 2026
Last updated: Jun 11, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.325
Output token cost: $3.25

Limits

Output tokens: 65,536 tokens
Context window: 256,000 tokens

Transparent token rates

Compare Qwen 3.6 27B pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Qwen 3.6 27B

Venice AI

CoverageBenchmark

This July 14, 2026 third-party deep dive on Qwen 3.6 27B compiles community-measured numbers rather than an official Alibaba report, citing an NVIDIA Hugging Face model card dated June 26, 2026 that describes the model as a 27-billion-parameter dense transformer with hybrid attention (Gated DeltaNet and Gated Attention The author is explicit that no official Alibaba benchmark report for Qwen 3.6 27B appears in the surveyed signal set, and every figure is sourced from named community testers running the weights locally. The article also flags that NVFP4 quantizations of the 35B sibling appear broken in agentic workflows, while the 27B

Venice AI

Official sourceRelease Notes

Venice's official changelog for April 21 – May 5, 2026 explicitly lists Qwen 3.6 27B among newly added text models on the platform, describing it as a "Text model from Alibaba Cloud." This is the strongest first-party confirmation of the model's existence and its arrival on Venice during that window, alongside other ad The same changelog post pairs the Qwen 3.6 27B launch with several other Venice product updates of the period, including Kling 4K video, a Voice Mode for realtime chat, and revised programmatic VVV burn amounts for new subscriptions. That context frames Qwen 3.6 27B as part of a broader model expansion aimed at develop

Venice AI

Coverage

This Hacker News thread, anchored to a post about Qwen3.8-Max, contains multiple practitioner comments that explicitly name the Qwen3.6-27B and Qwen3.6-35B variants. One commenter reports running Qwen-3.6-35B-A3B as a "workhorse" on a local Mac, an AMD R9700, and an RTX 5090, using it with OpenCode as the gateway into The same thread surfaces concrete qualitative limitations relevant to Qwen 3.6 27B: the bulk-task commenter tried a dozen quant and context-length combinations and concluded the 27B/35B outputs were too unreliable for code generation beyond easy tasks, recommending DS-V4-Flash quants for local coding work instead. A fo

Venice AI

Official sourceRelease Notes

Venice's aggregated changelog index records that a quantized "Qwen 3.6 27B FP8" variant was added on July 7, 2026 as a Text model running with E2EE/TEE privacy and made available to all users. This confirms an additional, inference-optimized variant of the Qwen 3.6 27B line beyond the original launch, signaling continu The same index also lists a sibling "Qwen 3.6 35B A3B" entry on July 20, 2026 under Private privacy, indicating that the Qwen 3.6 family on Venice spans multiple parameter counts and quantization profiles. Together with the FP8 variant, these entries show a deliberate rollout pattern around mid-2026, expanding privacy