Alibaba · Shipped
Qwen3.8-27B release date
Specs, benchmarks and changelog are on the Qwen3.8-27B model card at TPS.
First uncensored version on Hugging Face: JonathanColetti/Qwen3.8-27B-Uncensored-GGUF, 9 days after release. 279 uncensored versions in total.
Released
Aug 5, 2026
What we know
Scores 52 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Luna and outperforming much larger models.
Spec
- Params
- 27B
- Active
- Not stated
- Context
- 262144
- Weights
- Apache 2
- Modality
- Not stated
- Codename
- Not stated
Every claim
7 from 5 sources- Simon WillisonSpotted in the wildOct 4
Qwen3.8 27B model tested on addition-in-words task with medium reasoning enabled.
- Hacker News (AI front page)Spotted in the wildSep 18
Qwen3.8-27B model with 13.1 GB VRAM requirement documented in blog post.
- TechmemePressAug 17
Qwen 3.8 27B, a 27B parameter open-weight vision-capable model, was released with long context, tool calling, vision, and code generation capabilities.
- Hacker News (AI front page)Spotted in the wildAug 16
Qwen 3.8 27B model exists and is noted for defaulting to overthinking in its responses.
- TechmemeFrom the labAug 14
Alibaba released open weights for Qwen3.8-27B under Apache 2.0 license, claiming it outperforms Qwen3.7-Plus in coding and office workflows.
- OpenRouter catalogSpotted in the wildAug 14
Qwen3.8 27B is now available and routable on OpenRouter as an open-weight model.
- QwenSpotted in the wildAug 5
Qwen3.8-27B model repository published on Hugging Face, confirming model's existence and availability.