Authoritative benchmarks across 7 official boards, live API price drop alerts, and battle-tested .cursorrules / CLAUDE.md rule packages.
Top tier reasoning & autonomous refactor
Compare 29+ plans with starter credits
Key takeaways & expert breakdowns
Generate & export battle-tested prompt rules
Claude Code Changelog (@ClaudeCodeLog) announced Claude Code 2.1.272: the official changelog lists bug fixes and reliability improvements aimed at fewer crashes and more stable common workflows, shipping hours after the feature-heavy 2.1.271.
CoreSpeed (@CoreSpeedHQ) launched: connect apps through one MCP endpoint with built-in cross-agent memory, media generation, web and social research tools, plus agent identities, budgets, and activity logs. Start at https://corespeed.io.
Vercel CEO Guillermo Rauch (@rauchg) shipped coding-agent CLI fx 0.0.10: changelog cites up to ~1.6× faster turn completion and ~2.5× faster model-request start (biggest gains on long sessions), plus auto-upgrades, ctrl+g restart+resume, and stronger secret protection on MCP errors. Details: https://fx.sh/changelog#v0.0.10.
Swarms (@swarms_corp) launched Swarms Marketplace MCP: with a single MCP connection, agents can search, discover, use, publish, and monetize 6,000+ agents, prompts, and tools across the marketplace—aimed at MCP-capable coding setups such as Claude Code and Cursor.
OpenAI voice lead Atty Eleti (@athyuttamre) said ChatGPT Voice in the desktop app is ~60% cheaper for Codex and Work, yielding ~2.4x more voice time so you can orchestrate longer without burning the agentic pool as fast—and leave more tokens for real work.
Prasenjit Sarkar (@stretchcloud) covered Motion MCP: coding agents call Motion animation and video tools over standard MCP. In the demo, Codex produced a launch video from a system prompt for about $5 in about 10 minutes—so the same agent that ships code can also produce launch creatives without designer round-trips.
Claude Code Changelog (@ClaudeCodeLog) announced Claude Code 2.1.271 with 96 CLI changes. Highlights: Remote honors host fast-mode or /fast when the org allows it; Bash/PowerShell/Monitor auto-sandboxing can limit egress with per-command allowed_domains; switching account or API key clears the org-policy cache so rules refresh immediately.
@TiagerBao covered Cloudflare's open-source security-audit-skill (MIT) for coding agents: a six-phase flow that maps architecture and trust boundaries, assigns independent checkers, adversarially disproves candidates, emits findings.json, re-validates, and reports—labeling results confirmed, needs_validation, or rejected. Repo: github.com/cloudflare/security-audit-skill; works with Claude Code, Cursor, and Codex.
Mitchell Hashimoto (@mitchellh) demoed the Superlogical Rex multiplexer CLI: everything in the GUI is scriptable, plus more—create sessions, run commands, split panes, move focus, simulate key/mouse input, and wait on scripts. A streaming event system lets any client hook in, and a terminal API returns running-process JSON for editors and agentic coding tools.
Vercel engineer Chris Tate (@ctatedev) announced Labs is now an official Vercel team exploring primitives for developers and agents in the AI era—shipping agent-browser, skills, json-render, deepsec, fx, just-bash, portless, vgpu, and more. Support splits into Labs products (maintained), active experiments (fast, breaking), and past experiments; start at vercel.com/labs.
MiaAI Lab (@MiaAI_lab) added a TP=3 path to GLM-5.3-Flash EXL3 for three DGX Sparks: +28% decode and +9% prefill vs TP=2, with 3.2M KV cache and 1M context, plus optional NFS weight sharing from the head node (NFS_SHARE=1). Repo: github.com/MiaAI-Lab/GLM-5.3-Flash-EXL3-2x-DGX-Sparks.
Arena.ai (@arena) reports Agent Arena results: DeepSeek-V4.1-Flash (Max) posts about +4.87% net improvement at roughly $0.07 median task cost — lowest among the top-3 open models — retaining ~98% of Hy4 preview’s net gain at ~73% lower cost and ~76% of Kimi K3 (Max) performance at ~92% lower cost.
Inferact (@inferact) announced a Google Cloud partnership to make TPU first-class in vLLM: one engineering roadmap for production serving/kernels, a native PyTorch path via TorchTPU, and day-0 frontier-model support, plus a community program with shared TPU capacity and maintainer review — all open source.
Nous Research (@NousResearch) opened Hermes Agent for business: Hermes Business on Nous Portal lets teams share one balance with per-member spend caps and a shared skills library; Hermes Enterprise brings the same self-improving agent stack on-prem or in your cloud (SSO/SLAs). Start at portal.nousresearch.com/business.
OpenAI Developers (@OpenAIDevs) announced official Arch Linux support for the Codex app: install with the Arch installer and keep it updated via pacman, with no community repackaging. Docs: learn.chatgpt.com Linux app page.
A developer amplified Microsoft’s open-source tgrep (github.com/microsoft/tgrep): trigram-indexed regex search for large codebases with a client/server design and file watcher for millisecond lookups; the post says Copilot CLI already uses it internally to cut agent repo-scan cost.
GitHub staffer Pierce Boggan (@pierceboggan) announced Copilot Auto model selection now offers three tiers—efficiency (cost-first), balance (cost/quality/latency), and intelligence (quality-first). Auto still picks a model per prompt; usage is billed on the selected model, and paid subscribers keep a 10% Auto usage discount. Rolling out in VS Code, Copilot CLI, and the GitHub Copilot app (changelog 2026-09-14).
Official @ggml_org shipped llama.cpp v0.4.1: adds Maple 20B-A1B, Tencent Hy 4, and Spark2.5 support; improves Kimi-K3 rollback, JSON schema handling, qwen3-coder complex-type parsing, and MCP/multimodal fixes; updates ggml to v0.24.0. Release notes: ggml-org/llama.cpp v0.4.1.
Patrick Erichsen (@Pat_Erichsen) announced OpenClaw’s new Suggested Task feature: when a coding agent spots well-scoped work, it can recommend kicking off a new session so ambitious multi-agent work is delegated instead of stuffed into one thread.
Claude Code lead Boris Cherny (@bcherny) said Claude Mods are landing now. Community demos already include a Tetris-in-Claude mod; latest updates, technical details, and more demos are on anthropics/claude-code GitHub issue #91870.
