Moonshot AI released open-weight Kimi K2.6 on Kimi.ai, the API, Kimi Code, and Hugging Face, aimed at long-horizon coding (demos hit 4,000+ tool calls over 12+ hours) and agent swarms that scale to 300 parallel sub-agents across 4,000 coordinated steps.
Cursor says harness changes—trimming 66% of the system prompt, loading rarely used built in tools on demand (the same approach that cut MCP session tokens 46.9%), tighter prompt caching, and numbering only every tenth line—cut user token costs by 7% with no quality drop…
Google’s Antigravity SDK now runs agentic workflows fully offline with LiteRT and Gemma 4 26B A4B (plus Ollama, LM Studio, or vLLM via OpenAI compatible backends). A hybrid demo keeps code on device: Gemini 3.8 Flash plans in the cloud while local Gemma instances audit and patch, with 97% of tokens staying offline…
Google launched Gemini 3.8 Flash TTS and Flash Lite TTS on Sept 23: design voices from natural language prompts, pick from 2,000+ presets, or clone from a 30 second sample with consent checks, SynthID, and C2PA. Available in AI Studio and the Gemini API for voice agents, audiobooks, and dual speaker scenes…
Anthropic’s Opus 5.5 is cheaper than Opus 5, but its migration guide flags four breaking changes—always on thinking, no forced tool choice, append only thinking blocks, and a new computer use toolset—that can return 400s in existing agent loops. Claude Code and Managed Agents already handle it; custom agents need code changes before flipping the model ID…
Anthropic moved Claude Code cloud sessions (formerly Claude Code on the web) out of research preview on Sept 23 for Pro, Max, and Team, plus eligible Enterprise seats. Sessions run on Anthropic managed VMs so work continues with the laptop closed, sharing rate limits with other Claude Code usage…
Spotted something we missed? Start a thread.