Qwen3.8 27B

Qwen3.8-27B is a 27-billion-parameter dense post-trained model released by the Qwen team as part of the Qwen3.8 generation. According to the official model card, it is built on the architectural foundation of Qwen3.5, following the Qwen3.5 and Qwen3.6 series, and is positioned as the most capable generation in the Qwen open-model family to date. The model card was published on the official Qwen Hugging Face repository, where weights and configuration files are distributed in the Hugging Face Transformers format with compatibility for inference stacks such as Transformers, vLLM, SGLang, and TokenSpeed.

The model is described as a native vision-language model designed to understand images, with the Qwen3.8 generation emphasizing substantial gains across coding, professional work, research, and long-horizon agentic tasks. The model card highlights stronger autonomous planning and improved handling of environment feedback for more reliable end-to-end task completion, as well as broader support for popular development harnesses and tools. For users who prefer a managed deployment, the Qwen team announced a forthcoming hosted Qwen3.8-27B service through Qwen Cloud with production features such as a 1M-token default context length and built-in tools, though at the time of the card's publication this hosted service was still listed as coming soon.

Vultrqwen3.8-27bqwen

Open provider page Provider docs

Quick Info

Provider: Vultr
Model key: qwen3.8-27b
Release date: Aug 14, 2026
Last updated: Aug 14, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.15
Output token cost: $1.00

Limits

Output tokens: 32,768 tokens
Context window: 262,144 tokens

Transparent token rates

Compare Qwen3.8 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 Qwen3.8 27B

Vultr

CoverageBenchmark

Alibaba's Qwen team released Qwen3.8-27B on August 14, 2026 as a 27.78-billion-parameter dense multimodal checkpoint shipping under Apache 2.0. The model accepts text, images, and video and offers a native 262,144-token context window. Qwen reports substantial gains over Qwen3.6-27B in agentic coding, computer use, and vision-language tasks without increasing the published decoder size. Qwen's official model card shows Terminal-Bench 2.1 rising from 63.4 to 73.0, DeepSWE 1.1 from 13.3 to 42.2, OSWorld-Verified from 63.9 to 84.3, and SWE-MM from 25.7 to 38.6 versus Qwen3.6-27B. Kingy.ai's launch-day review cautioned that the scores are vendor-reported, several benchmarks are in-house or modified, and one-million-token contexts require YaRN scaling rather than the native window. Kingy found no independent benchmark reproduction on launch day.

Vultr

CoverageAnalysis

Hacker News discussion highlighted that Qwen3.8 27B achieved a score of 52 on the Artificial Analysis Intelligence Index, a substantial jump from Qwen3.6 27B's score of 38, which had been the highest in the small open-weights category. Commenters noted the new model now matches DeepSeek V4 Flash 0731 and beats all medium-category open-weight models. The 27B size makes it attractive for hardware-constrained users with plenty of compute but limited RAM. Participants flagged that the 52 score likely reflects the Max reasoning variant with very long reasoning traces, citing roughly 2.3x the token usage of GPT Luna Max. The HN thread observed Qwen3.8 27B is comparatively slow and uses more tokens per task than peers like Gemma4, due in part to the xhigh default reasoning mode. Output speed on VLLM was reported around 120-180 TPS for the Qwen 3.6 35B-A3B sibling.