Qwen3.8 27B

Qubrid AI compiled official and independent benchmarks for Qwen3.8-27B on August 31, 2026. The model scores 52 on the Artificial Analysis Intelligence Index at maximum reasoning effort and 61.7 on SWE-bench Pro per Qwen's own evaluation, leading its comparison set on agentic coding and computer-use benchmarks while tra The generational jump on agentic coding is highlighted: DeepSWE 1.1 moves from 13.3 to 42.2 and QwenSWEBench from 49.3 to 79.0 versus the previous 27B, and the 27B beats closed-weight Qwen3.7-Plus on most agentic and coding rows while trailing it on Humanity's Last Exam and GPQA Diamond. Vision-language scores include

Kingy.ai published a launch-day review of Alibaba's Qwen3.8-27B on August 14, 2026, characterizing it as a strong candidate for the best dense, locally deployable multimodal model around 30 billion parameters. The checkpoint contains 27.78 billion parameters, accepts text, images, and video, ships under Apache 2.0, and Kingy.ai also noted caveats: launch scores come from Qwen, several benchmarks are in-house or modified, the SWE-bench Pro comparison imports Anthropic's Opus result rather than rerunning it, and claims about local hardware ignore KV cache since one-million-token contexts require YaRN scaling that Qwen warns can hurt sh

The NVIDIA Developer Forums thread (dated August 8, 2026) carried an announcement linking to the qwen.ai blog for Qwen3.8-27B, confirming that the 27B variant would release as open weights the following week, alongside the larger Qwen3.8-Max flagship. The thread acts as a community index of closely related sub-topics o Although the thread itself predates the August 14 release and is anticipation-focused rather than a primary announcement, it usefully documents the open-weights rollout context for Qwen3.8-27B and surfaces community-quantization threads (NVFP4 and MixedInt4-AutoRound variants) that target single-GPU DGX Spark deploymen

Released August 14, 2026 by Alibaba Cloud's Tongyi Lab under Apache 2.0, Qwen3.8-27B is a 27.78-billion-parameter dense multimodal model positioned for local deployment on a minimum of 24GB VRAM. The analysis details a 64-block transformer (hidden dim 5,120; 24 query / 4 KV heads; head dim 256; vocab 248,320), a native The write-up frames Qwen3.8-27B as outperforming Meta's "Muse Glimmer (30B)" on all eight direct comparison benchmarks and surpassing Claude Opus 4.6 on 15 of 19 overlapping tests, presenting the release as demonstrating that architectural refinements like Gated DeltaNet and multi-token prediction can close the dense-v