Alibaba Open-Sources Native Multimodal Qwen3.8-27B Model

Alibaba's Qwen team has open-sourced the Qwen3.8-27B, a native multimodal dense model with 27B parameters under the Apache 2.0 license. The team claims it comprehensively outperforms Qwen3.7-Plus, excelling in coding and office workflows, supports 262K native context extendable to 1M, and quickly topped the Hugging Face trending charts. The community has released various quantization options for local deployment, and third-party frameworks like SGLang and Ollama have rapidly integrated support.

Confirmed

• Model Architecture and Performance: Qwen3.8-27B is a 27B parameter dense hybrid attention model (48/64 layers use linear attention), natively supporting image-to-text tasks. It supports 262K native context, extendable to 1M via YaRN, includes a built-in MTP draft head, and is supported by vLLM 0.17.0.
• Deployment and Quantization: The official FP8 quantized version is available on Hugging Face. Unsloth released GGUF and NVFP4 quantized weights, with the NVFP4 version also trending on HF; Ollama announced integration, supporting tools like Claude Code and OpenCode with optimizations for Apple Silicon.
• Inference Support: SGLang achieved 200+ tok/s decoding speeds on RTX 5090 and RTX Pro 6000 using NVFP4 quantization, and it is also deployable on DGX Spark.
• Community Feedback: The initial model card lacked benchmark results, but @Tccybo subsequently compiled the scores into comparison charts.

Unconfirmed

• Performance Comparisons: Community opinions suggest it rivals Claude Opus 4.6. @kimmonismus stated it surpasses Claude Opus 4.6 Max in agentic coding and computer use benchmarks; @de4dee relayed user views comparing it to Opus 4.6, though officials have not made this claim directly.
• Hardware Expectations: @OddTumbleweed574 noted it as the first sub-30B parameter model runnable on 3090 or M4 Pro, predicting a shift of more workflows to local runs by the end of the year.

Timeline

• Aug 13: Users spotted the model on ModelScope.
• Aug 14: Qwen officially released the FP8 quantized version on Hugging Face; the model card was published but benchmarks were pending.
• Aug 15: Benchmark scores were released, community summary charts were posted, and Ollama and SGLang announced integration or support.

2026-08-13 ~ 2026-08-15 · 48 related posts

Full story(3 episodes)→

• Episode 1: Qwen3.8-27B Model Confirmed for September 3 Release(2026-08-11, 2 posts)
• Episode 2: Alibaba Teases Qwen3.8-27B with Unmatched Intelligence Density(2026-08-12, 11 posts)
• Episode 3: Alibaba Open-Sources Native Multimodal Qwen3.8-27B Model(2026-08-13, 48 posts)

• Qwen3.8-27B Model Surfaces on ModelScope Ahead of Hype — Ok-Shower7286 · 2026-08-13
• Qwen releases Qwen3.8-27B-FP8 quantized model with image-text-to-text support — Qwen · 2026-08-13
• Rumor: Alibaba's Qwen 3.8 27B Model Set for Friday Launch — Brilliant-Hall1387 · 2026-08-14
• [source] Qwen3.8-27B Model Card Goes Live, Benchmarks Pending — -Cubie- · 2026-08-14
• Qwen3.8-27B FP8 Weights Released on Hugging Face — Certain-Cod-1404 · 2026-08-14
• Qwen 3.8 27B released: best local dense model yet — erdaltoprak · 2026-08-14
• Qwen3.8-27B Released, Claimed to Match Opus 4.6 — de4dee · 2026-08-14
• Qwen3.8-27B: 27B params rival frontier, beats Claude Opus 4.6 Max on agent benchmarks? — kimmonismus · 2026-08-14
• Alibaba Releases Qwen3.8-27B: 27B-Parameter Multimodal Model Beats Claude Opus 4.6 Max on Several Benchmarks — kimmonismus · 2026-08-14
• [source] Alibaba open-sources Qwen3.8-27B: 27B multimodal model beats Qwen3.7-Plus, Apache 2.0 — Alibaba_Qwen · 2026-08-14
• Alibaba Qwen Open Sources 32B Native Multimodal Model with 1M Context Support — 千问大模型 · 2026-08-14
• Unsloth Releases Qwen3.8-27B GGUF Weights for Easy Local Deployment — kevin_1994 · 2026-08-14
• Qwen3.8-27B Open-Sourced: 27B Params Outperform Qwen3.7-Plus, Native Multimodal — NVIDIAAI · 2026-08-14
• Unsloth releases Qwen3.8-27B quantized: NVFP4 1.5x faster, retains 92-97% accuracy — danielhanchen · 2026-08-14
• Qwen3.8-27B Released: Checkpoint Loved by Startups and Single-Node Users — tokenbender · 2026-08-14
• Qwen3.8-27B tops Hugging Face trending, Apache 2.0 open weights — Qwen · 2026-08-14
• Qwen 3.8 27B GGUF Quantized Version Trending on Hugging Face — unsloth · 2026-08-14
• Qwen3.8-27B Released, Hugging Face Collection Available — song91 · 2026-08-14
• Qwen3.8-27B Released: Coding, Agentic, General, and Vision Benchmarks Shown — multimodalart · 2026-08-14
• Qwen3.8-27B Becomes 4th Most Liked Model on Hugging Face of All Time — multimodalart · 2026-08-14
• Qwen3.8-27B open-sourced with SGLang Day-0 support, 206 tok/s on RTX 5090 — ying11231 · 2026-08-14
• [source] Qwen3.8-27B launches with 1M context on a single GPU, vLLM ready — Alibaba_Qwen · 2026-08-14
• Alibaba Qwen Open Sources Qwen3.8-27B, Outperforms Claude on Benchmarks — 智东西 · 2026-08-14
• Alibaba releases Qwen3.8: 27B open-source model with 262K context — ProfBuehlerMIT · 2026-08-15
• Qwen3.8 benchmark charts visualize performance across tests — Tccybo · 2026-08-15
• Qwen3.8-27B launches with day-zero AMD support for local AI development — Alibaba_Qwen · 2026-08-15
• HF Likes vs Downloads: EleutherAI Models 125M Downloads, <2k Likes — BlancheMinerva · 2026-08-15
• Qwen Official: Qwen3.8-27B Significant Jump, Try It Out — Alibaba_Qwen · 2026-08-15
• Unsloth releases NVFP4 quantized Qwen3.8 27B for efficient local deployment — airesearch12 · 2026-08-15
• Qwen 3.8 27B Released: Runs on a Single GPU, Beats Opus 4.6 Max on Several Benchmarks — cephaloform · 2026-08-15
• Qwen3.8-27B Runs Locally: 17GB RAM with Unsloth Dynamic GGUFs — danielhanchen · 2026-08-15
• Alibaba Releases Qwen3.8-27B: Apache 2.0 Local Model — adrianscottcom · 2026-08-15
• unsloth Releases Qwen3.8-27B NVFP4 Quantized Model, Trending on Hugging Face — unsloth · 2026-08-15
• Qwen 3.8 27B Now on Ollama with Tool Integrations — ollama · 2026-08-15
• Ollama adds Qwen3.8 27B, one of the best open models for agentic tasks — ollama · 2026-08-15
• SGLang adds day-0 support for Qwen3.8-27B, achieving 200+ tok/s on RTX 5090 — unseenmarscai · 2026-08-15
• Alibaba releases open weights for new Qwen3.8-27B model — Polymarket · 2026-08-15
• Qwen 3.8 27B Released: Local AI Alternative, May Replace Cloud Subscriptions — Odd_Tumbleweed574 · 2026-08-15
• Alibaba releases 27B open-weight agentic model Qwen 3.8 — markjeffrey · 2026-08-15
• Alibaba Releases Qwen3.8-27B Model on Hugging Face — _akhaliq · 2026-08-15
• Alibaba releases Qwen3.8-27B with 11 finetunes and 145 quants — victormustar · 2026-08-15
• Qwen 3.8 27B released: official and community quantized versions roundup — sammcj · 2026-08-15

6 near-duplicate retellings: Alibaba_Qwen · cedric_chee · haider1 · airesearch12 · FlolightC · gajesh