A DeepSeek Harness team member explains the framework's plugin philosophy: it will ship stable core capabilities and common built-in features, but under an 'eve…
OpenAI's Tibo clarifies that all Pro plans have no 5-hour limits and offer 10X, 20X, and 50X usage versus Plus, with Pro 500 including ultrafast access.
OpenCode's dax summarizes the past week's releases: Hex for talking to agents, Browse for agent browser use, Rune for native apps, a DeepSeek $100-for-$10 deal,…
A dot upgrade lets it use sign-in codes sent to your connected inbox to log in on your behalf, with one-time or always-allow options when prompted.
A user benchmark shows OpenCode Browse running 40% faster than Vercel's agent-browser with 81% fewer input tokens and 33% fewer steps, using GPT 6 Astra and no…
A user shares pricing for GB300 and B300 racks: GB300 at $9-10/hr with under one-year commitment and one-week readiness, B300 at $6-7/hr on the same terms.
A developer reports that AI successfully ported a TUI app to the Rune framework with good results, surprising the project's creator given it launched only a wee…
Ahmad Awais says CommandCodeAI's internal RSI engine uses cache as its north star metric, enabling $100 worth of inference for $10 and surfacing 2-3M code impro…
OpenCode Browse now supports running the agent and the controlled browsers on separate machines, so a remote server can orchestrate browsers on a desktop and la…
LiteLLM's weekly update covers a full migration of routing to Rust, new model support, a Codex model picker, -latest model aliases, and the Skills Hub.
