Cloudflare announces Clef decision models and fine-tuning service
Cloudflare announced the Clef and Clef-flash decision models on Workers AI, publishing their weights on Hugging Face under the Apache 2.0 license. The company claims Clef outperforms competitors on some benchmarks. Fine-tuning will initially be offered with engineering support; a self-service platform is planned for later.
Cloudflare released its hosted Clef and Clef-flash decision models on Workers AI with open weights, and announced a fine-tuning service for Clef.
02INFRASTRUCTURE
turbopuffer makes vector indexing secondary in v3
In v3, turbopuffer changes its storage layout, demoting the ANN index from a primary key to a secondary index. The company aims to speed up text, regex, and vector searches and support GROUP BY queries. It plans to reach performance parity before putting the release into production.
turbopuffer v3 aims to establish a new storage architecture for text, regex, and vector searches by making the ANN index secondary.
03DEVELOPER TOOLS
Rust compiler runtime fell 4.57% on average in two months
From July 29 to September 28, 2026, 555 of the Rust compiler’s 629 benchmark measurements improved, with average wall-clock time falling 4.57%. The LLVM 23 upgrade alone cut the average by 1.2%. The new borrow checker and trait solver slow compilation in some cases.
Recent changes to the Rust compiler improved 555 of 629 benchmark measurements, reducing average runtime by 4.57%.
04AI
OpenAI and Synopsys to develop a chip-design model
OpenAI and Synopsys formed a multi-year partnership to develop GPT-Synopsys, a chip-design model for Synopsys’ electronic design automation tools. The companies aim for the model to run tools, interpret results, and iterate on designs. Early customer work is underway, but development and expected benefits remain unsettled.
According to the announcement, the service will run on OpenAI infrastructure; the companies say customer design data will not be used to train models.
05AI
Language models manage context like a file
The paper introduces Context Language Models (CLMs), which manage context as a file and let the model update it freely. The authors report that CLMs outperform context-management strategies on various tasks; on BrowseComp-Plus, they achieve 11.4% higher accuracy with 21.5% fewer FLOPs.
The new CLM approach lets a language model manage its context as a file and update it directly.
06DEVELOPER TOOLS
Bez aims to generate a web engine from specs and tests
In the project’s current state, DOM, style, and tree structures are handwritten, while models generated eight of nine CSS 2.1 layout rules. These rules were tested with 227 examples and 11 usable WPT normal-flow pages; the engine is not yet comprehensive.
The Bez project aims to implement web-engine rules by generating code from specifications and browser tests.
07DEVELOPER TOOLS
Janus: a Go tool for running GGUF models locally
Janus aims to run GGUF models locally on a CPU or Vulkan and provide an OpenAI-compatible API as a single Go-built binary. According to the project description, it uses llama.cpp; the Windows build downloads Vulkan DLLs, while Linux requires libllama.so.
Janus is presented as a Go-based tool that runs GGUF models on a local CPU or Vulkan and provides an OpenAI-compatible API.
08DEVELOPER TOOLS
kcc, a C17 compiler written in ARM64 assembly
The project says the compiler checks and optimizes C17 code, converts it to native machine code, compiles itself into the same binary, and cross-compiles for x86-64. It also says it has compiled Lua, SQLite, and DOOM, and booted the Linux kernel.
kcc is a C17 compiler developed as a single literature-based program that can compile its own source code.
✓
That’s it for today.The rest of your time is yours.
Your daily technology digest, every morning in English. Free. Unsubscribe anytime.
Sources, in the open.
Stories are selected from Hacker News; accessible original articles are summarized and translated with AI. Summaries may contain errors; check the source beneath each story for details. Preparation starts daily at 06:55 Türkiye time (UTC+3), targeting publication before 07:00.
Stories are read aloud by an AI voice.
This edition covers articles shared on Hacker News on the date shown above.
Now on the App Store for iPhone
Your daily digest. Now in your pocket.
Read or listen to a new edition every morning with Kısa for iPhone, save stories for later and keep your streak.
