Matching stories Newest first
Simon Willison describes Battle Drones, a vibe coded RTS built with Claude
Willison built the browser RTS mostly with coding agents, mainly Claude Code, over 12 iterations and 321 commits. It includes audio-based multiplayer for two devices in microphone range, and he says he reviewed almost none of the code.
Hugging Face model Spark-X2.6-4B, a compact on-device language model, is released
Spark-X2.6-4B keeps the Spark-X2.5-4B architecture and gains from improved post-training, with three thinking levels (none, low, high). The developer claims it matches or beats models around 9B parameters on open-source benchmarks.
Google releases a PyTorch developer tool to move apps off Nvidia GPUs
Google follows Amazon with a developer tool that helps PyTorch users shift their apps off Nvidia GPUs.
Six-way benchmark compares Qwen3.8-Flash-Next locally against Claude Opus 5.5 on a Rust feature
A community test of 12 PRs on a LlamaStash feature found Opus 5.5 averaged 82% against 68% for Claude Code with Flash-Next and 66% for Pi with Flash-Next. Flash-Next ran locally at about 4.16 bpw, so the author paid only electricity, and results are a single developer's unverified experiment.
DeepSeek open-sources DeepInfer, the serving stack it runs in production
It ships under Apache 2.0 and for now supports only DeepSeek-V4.1-Flash in offline mode. It leads Hacker News with 1,862 points.
Hackers breach OpenCode AI coding tool and steal user API keys
Attackers got into part of the open source coding agent's online platform, stole some users' API keys, and ran up charges on their AI accounts. AI tools the team consulted refused to analyze the malicious code.
OpenAI publishes coding evaluations for low-resource programming languages
OpenAI says it is improving ChatGPT on low-resource programming languages, with new evaluations covering software engineering, SQLite and COBOL migration.