Qwen3.8-27B Now Operates at 70 Tokens/Second on Laptops
Qwen3.8-27B can now process 70 tokens per second on laptops thanks to the new DFlash 2 technology. This significant speed improvement comes without sacrificing output quality.
Qwen3.8-27B Now Operates at 70 Tokens/Second on Laptops
The new DFlash 2 has been released, representing a new generation of speculative decoding based on block diffusion.
DFlash 2 accelerates Qwen3.8-27B to 70 tokens/second on a MacBook Pro with M5 Max. The acceleration of autoregressive decoding reaches 4.6 times, while the final output remains completely identical to the original model — literally not a single differing token.
New Features
New draft models for Qwen3.8-27B are already available with native support for SGLang, vLLM, llama.cpp, and oMLX.
Essence of the Approach
Learn more about the approach at: DFlash 2
Why it matters
AnalysisThis is important because new technologies like DFlash 2 significantly enhance the performance of AI models, allowing them to be used on more accessible devices. This opens up new opportunities for developers and researchers in the field of artificial intelligence.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
JSON — Structured Data for Programs, APIs, and AI Agents
A complete hands-on guide to JSON: syntax specifications, serialization via parse and stringify, REST API consumption, JSON Schema validation, and structured outputs for LLMs.
DOCX and XLSX Formats: When AI Needs to Deliver Production-Ready Documents
A comprehensive hands-on guide to generating and processing DOCX documents and XLSX workbooks with AI: comparing against Markdown and CSV, prompt engineering, formula design, and code automation.
Binance publishes investigation into $6.3 billion theft
Binance’s investigation links the theft to a modified Homebrew download, a fake interview and abuse of an internal AI tool.
Reddit user explains why Claude Code weekly limits are burning almost 4x faster
A Reddit user compared token usage across two weeks and found that each 1% of Claude Code's weekly limit now yields far fewer output tokens, pointing to subagents, reasoning time, and UI refresh requests as likely causes.