Google’s Sept 17 Android Bench 2.0 adds multi-day long-horizon tasks and continuous scoring; full-task pass rates top out around 28% (vs ~91% on the old incremental set). GPT-6 Astra leads the leaderboard; the suite also scores Gemini Flash, Fable 5.1, Kimi K3, and Qwen, with agent runs on Codex and Antigravity.
Shown on HN Nov 2, the Kyutai Pocket TTS fine tune takes tags like [laugh] , [sigh] and <whisper inside the text, clones a voice from a few seconds of audio, and the author says it runs about 20x real time on four Apple M4 CPU threads from one 112 MB file. Weights are CC BY 4.0, it's English only, and the author notes tags vary by seed and no WER benchmark was run on this checkpoint…
Shown on HN, it drives WKWebView, WebKitGTK or WebView2 with no Node.js, and claude mcp add bruster bruster mcp quiet turns navigate, snapshot, click, fill form and screenshot into tools for Claude Code, Cursor, Codex or OpenCode. Input uses synthetic DOM events, cross origin iframes are out of reach, and Linux hidden mode still needs a display or Xvfb (Show HN)…
Shown on HN Nov 2 under MIT, pipx install arvre gives a TUI that reads /.claude/projects , links each fork to the earlier session sharing the most message IDs (Claude Code records no parent link), and lets you jump between parents and children, then resume one with claude resume or copy the command. It needs Python 3.11+ on macOS or Linux, and the author warns the session format is undocumented, so a Claude Code update could break it…
Shown on HN Nov 2, one pasted line connects your coding assistant to 3,257 wiki entries, 113 blueprints and 45 code recipes for things like jump feel, loot odds and saves, plus checkers that walk every story path to find dead ends and stuck quests while it versions your project. It's free to start with a Google or email sign in, in English and German, and the site gives no pricing beyond that…
Shown on HN Nov 2 under Apache 2.0, the alpha operator lets OpenAI run the agent loop while codex exec server executes commands in a Kubernetes Agent Sandbox workspace, with a gateway saving messages, results and approvals in PostgreSQL so follow ups reuse the same workspace. Agents, Environments and Profiles are CRDs, and a Profile can set an optional per session dollar cap; real tasks need Agents API beta access and incur API charges…
Spotted something we missed? Start a thread.
