Mixlayer shipped Qwen 3.8 27B on its new in-house speculative decoding backend written entirely in Rust.
Public source
Publisher name
Public post
A model launch that should have taken one day turned into a week of round-the-clock debugging. On paper, launching Qwen 3.8 27B was just a weight swap. But we decided to…
Company
Mixlayer
Frontier open-source models, at a fraction of the cost.
- Industry
- Technology, Information and Internet
- Location
- San Francisco, US
- Company size
- 2–10 employees
About Mixlayer
Frontier-grade APIs for open source AI models. We help you build incredible experiences and products on top of the best LLMs in the world at a fraction of the price.
See moreLatest activity
Latest activity from Mixlayer
5 signals
Discover more
Similar signals
Similar public activity from other companies.
Products & Services
Mirai Labs
Mirai Labs released a speculative decoding implementation in its inference engine Uzu for Qwen3.6 27B, Qwen3.8 27B, and Muse Glimmer.
Products & Services
Qwen
Qwen upgraded Qwen3.8-Max to include 2.4T parameters, 1M context tokens, and built for real world complexity.
Products & Services
SGLang
SGLang launched Day-0 support for the flagship K2-Horizon-375B model built for reasoning and agentic work.
Products & Services
Unsloth AI
Unsloth AI released Qwen3.8-27B GGUF, which reached 10M downloads and 3.7K likes on Hugging Face in 24 days.
Products & Services
TypeSafe AI