Keeping vLLM's Prefix Cache Warm Between Agent Turns From 55% to 95% cached 🔗 I’ve been playing with a few different ways to host Qwen3.8 locally. I’m aiming for something that can replace Claude Code for most of my tasks.
model
Qwen3.8-27B-FP8
huggingface.co/Qwen/Qwen3.8-27B-FP8 ↗
7671809 downloads787 likesimage-text-to-texttransformers
from the model card
Qwen3.8-27B-FP8 [!Note] This repository contains FP8-quantized model weights and configuration files for the post-trained model in the Hugging Face Transformers format. These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, TokenSpeed, etc. The quantization method is fine-grained fp8 quantization with block size of 128, and its performance metrics are nearly identical to those of the original model. [!Tip] For users seeking managed, scalable inference without infrastructure maintenance, the official Qwen API service is provided by Qwen Cloud. In particular, Qwen3.8-27B will be available as a hosted version with more production features, e.g., 1M context length by default, official built-in tools. For more information, please refer to the Qwen3.8-27B Overview. The service is coming soon. Stay tuned for updates. Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date. Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi…
discussions
- Qwen 3.8 15 ongoing since 2026-09-09
recent items
Keeping vLLM's Prefix Cache Warm Between Agent Turns (doug.sh via hn) Shapelearn Qwen 3.8 27B (13.1 GB VRAM) (byteshape.com via hn) ShapeLearn-Lite Held Up. ShapeLearn Did Better: Qwen 3.8 27B We were a little impatient.
Alibaba releases Qwen 3.8 Omni Flash (qwen.ai via hn) Qwen Studio offers comprehensive functionality spanning chatbot, image and video understanding, image generation, document processing, web search integration, tool utilization, and artifacts.
Ask HN: What is the best local model that runs on your Mac at decent speed? (news.ycombinator.com) What are your own experiences on this? What mac/specs do you have?
Show HN: Swift-Qwen3.8-27B, -58.3% thinking, x1.95 speed, accuracy of xhigh (huggingface.co via hn) Hi everybody, we post-trained Qwen 3.8 27B to be more efficient by figuring out which tokens were linked to overthinking and penalizing them without "attacking" the reasoning length directly then fixed the accuracy with a bit of secret sau…
Operating System powered by Qwen 3.8 27B at 1950 tokens/SEC (twitter.com via hn) Operating System powered by Qwen 3.8 27B at 1950 tokens/sec! here is what 1,950 tokens/second @Alibaba_Qwen's 3.8 27b actually looks like on @cerebras: i wrote a minimal python web server that turns cerebras inference into a live operatin…
Qwen3.8-Flash passed 17/17 checks on a DeepSWE v1.1 task in our agent harness (github.com via hn) AIC Evals Verified benchmark artifacts, model patches, and integrity receipts for AIC, a closed-source, host-enforced AI software engineering runtime. This repository publishes inspectable evidence from selected AIC evaluation runs.
Tuning a Local Coding Agent: Oh My Pi and Qwen3.8-27B on Two RTX 3090s (doug.sh via hn) Tuning a Local Coding Agent: Oh My Pi and Qwen3.8-27B on Two RTX 3090s Running a coding agent at home 🔗 I’ve started to use Oh My Pi , usually called omp, a fork of the open-source Pi coding agent, with Qwen3.8-27B, an open-weights model t…
Qwen 3.8 Flash via DwarfStar (twitter.com via hn) Qwen3.8 Flash Next is now supported in DwarfStar, covering 64GB Mac systems very well and with very fast inference of 50~70 t/s and > 1400 t/s prefill. For now this is Metal only.
GVS5H: Five Qwen3.8-27B Models Match Claude Fable 5 on LiveCodeBench Hard (github.com via hn) GVS5H: Five Qwen3.8-27B Models Match Claude Fable 5 on LiveCodeBench Hard Results Abstract. Frontier coding performance is typically bought with larger proprietary models at high cost.
Off-Target Effects of Response-Style Alignment in a Korean 27B Language Model (arxiv.org) We post-train Qwen3.8-27B for Korean response style -- verbosity, list and markdown usage, discourse structure and register -- and measure two behaviours the objective never targets: abstention on ambiguous social questions in KoBBQ, where…
Mining Qwen 3.8 reasoning trace for prompt/skill evaluation (olegivye.com via hn) I build tools for AI and LLM agents — orchestration, sandboxes, and the data pipelines that feed them — with a taste for what happens under the hood in Python, Rust, and Linux. My background is in Backend and AI/Data engineering, shaped by…
Qwen 3.8 follows GPT-5.5 Pro reasoning prefills (gist.github.com via hn) A follow-up to Reasoning prefills on a few open models and Stolen Thoughts This v1.1 reruns the reasoning-prefill experiment with GPT-5.5 Pro as the teacher. For each problem, I generated two responses from each target model: - an ordinary…
Peer-to-peer LLM inference in browser tabs, Qwen 3.8 27B (github.com via hn) SwarmLLM Every device brings a slice. Together they run the whole model.