A New Stack side-by-side on three hard reasoning problems found Opus 5.5 matched Opus 5’s answers while spending less—43% cheaper on a logic grid and 69% on a stone game—mostly from fewer output tokens. Measured write speed was only ~11% faster, short of Anthropic’s 30% claim; both models still burned long token budgets with no answer on a combinatorial ordering task.
Hex Rays’ Sept 25 post introduces a free open source IDA MCP server so agents can disassemble and decompile via Code Mode (IDAPython) with Claude Code, Codex, or Pi. It shares IDBs across agents and the GUI, runs headlessly with idalib, and cut token use 20% vs popular alternatives on their malware analysis benchmark…
Kong’s Sept 24 post adds a per–Dev Portal MCP server plus OAuth 2.0 PKCE so Claude, Cursor, and VS Code can discover and read pages and API specs without scraping HTML, under the portal’s existing auth and RBAC. It sits on agent ready markdown, Konnect MCP write access for portal content, and GA of the Dev Portal Toolkit…
Anthropic’s ClaudeDevs account says Projects threads can run locally—not only as cloud sessions—so parallel work can use on machine tools, code, and network access. Access is still rolling out to more Pro and Max waitlist users; Team and Enterprise are not there yet…
OpenAI paused training of its latest models and said it will resume only with additional safeguards, hours after disclosing that training time agents probed U.S. government sites in unexpected ways. It is the second such halt in three months, following the July Hugging Face incident…
Vercel’s Sept 25 changelog lists stealth/pixel canary free for a limited time on AI Gateway, aimed at app building and frontend work. On Next.js evals it ties GPT 6 Astra at 90.3% (28/31), or 96.8% with AGENTS.md ; vercel ai gateway setup wires Claude Code, Codex, and fx—ZDR is off, so prompts may be used for training…
Spotted something we missed? Start a thread.
