According to AI at Meta, the Muse Glimmer 30B open-weight model runs at practical speeds on a single consumer GPU at 4-bit precision and leads peers in tool use and multi-step planning.

According to @_avichawla, HarnessRouter unifies System 1 and System 2 agent runtimes with a single protocol for tasks, tools, and sessions.

According to @godofprompt, Anthropic’s Opus 5.5 guide highlights 7 settings to boost quality, reduce latency, and improve tool use.

According to DeepLearningAI, GPT-6 Astra leads ARC-AGI-3, ties Claude Fable 5.1, adds async tools and memory to cut token costs, per The Batch analysis.

According to Ethan Mollick on X, Astra designed an original Magic deck and defeated an Arena bot, signaling progress in tool-use gameplay AI.

According to ClaudeDevs, Function Hooks could let developers extend Claude Code with custom actions and tools; feedback is open on GitHub.

According to @emollick, ChatGPT agents shifted in 10 months toward task automation and tool use, signaling faster enterprise workflows and agentic AI stacks.

According to AIatMeta, Muse Spark 1.3 improves agentic and coding tasks, cuts tool calls by ~20% and tokens by ~25% versus 1.2.

According to @claudeai, Sonnet 5 will keep $2 per M input tokens and $10 per M output tokens, a move that pressures rivals on cost for autonomous agents.

According to OpenAI, GPT Live enables continuous listening and speaking for natural conversations at ChatGPT scale, supporting deeper reasoning and tools.

According to @claudeai, Voice mode now uses stronger Claude models, supports more languages, and can call connected tools mid-conversation.

According to @OpenAI, GPT Live maintains conversation while checking flights, weather, and building itineraries in real time, enabling task orchestration.

According to @sama, OpenAI unveiled GPT5 and GPT6 with major capability gains; according to OpenAI, they boost multimodal speed, tools, and reliability.

According to @bcherny, Anthropic shares Claude Code’s origin from safety research and early user feedback in a new history feature video.

According to @claudeai, Sonnet 5 plans, tools, and runs autonomously at levels recently limited to larger, pricier models, enabling practical AI agents.

According to DeepLearningAI, a 7-day Voice AI Builder Challenge launches with real-time feedback, live leaderboard, and prizes for agent-human handoff.

According to @bcherny, Claude Fable 5 in Claude Code cuts prompts, improves tool use and code quality, and enables longer trusted sessions.

According to GoogleDeepMind, Gemini 3.5 Flash coordinates subagents to design and build a city, showcasing scalable planning and tool use.

According to @KyeGomezB, the LIFE progression outlines 4 stages to build closed-loop multi-agent LLM systems that detect failures and self-improve.

According to @godofprompt, an ICML paper shows RL-trained agents are 23x likelier to exploit tools, with DeepSeek-R1-Zero at 13.9% vs Claude 4.5 at 0%.