OpenClaw adds Tencent Hy4 through TokenHub and TokenPlan
OpenClaw now lists Tencent Hy4, a vision model with a 1,024,000-token window, through TokenHub and TokenPlan.
OpenClaw now lists Tencent Hy4, a vision model with a 1,024,000-token window, through TokenHub and TokenPlan.
CrowdStrike is running Falcon Guardian inside Claude Cowork's VM, in beta on macOS, with Windows support expected later in November.
OpenAI will acquire Pierre Computer, whose Trees and Diffs already power Codex file browsing and change review, and work to bring Code Storage's features into Codex after close, pending customary conditions.
Claude Code 2.1.318 records answers to agent-written mod prompts, adds per-mod effort, and supports assume_role for AWS cost attribution.
OpenAI says ChatGPT and Codex are on CAD and simulation work, and that Astra designed a Codex device that still needed manufacturing review.
Google and Mercor release Physical-Engineering's Last Exam: 85 agent tasks in four disciplines, with a best hidden-set score of 48.3 percent.
langgraph-sdk 0.4.12 removes fairness_weight from PrioritySettings and rejoins a run stream that closes before the run finishes.
OpenAI's Pando Git backend now serves its monorepo, cutting median push latency nearly 5x under Codex-scale concurrent development.
Snowflake’s Expedition updates let CoCo and CoWork act on governed data, with cloud agents and scheduled automations for pipeline work.
Vercel MCP now renders live deployment status and build logs as an interactive card in ChatGPT and other MCP App clients.
OpenAI’s next internal security sprint pairs with Cathedral. A critical finding in a core product such as ChatGPT, Codex, or the API Platform can earn up to $30,000, and exceptional critical findings on those targets remain eligible for up to $100,000, or up to $172,000 with bonuses.
OpenAI's LifeVal eval tracks whether models can finish everyday personal-agent tasks, from childcare and housing to finances and government paperwork.
Hermes can import a Grok Bot through a self-written export, turning scheduled routines into paused cron jobs without copying credentials.
Claude Code 2.1.317 warns on unexpected schedule connectors, asks before sending your account email, and lets failed Agent SDK hooks block the action.
Elon Musk said Grok Bot’s serving cost was unsustainably high relative to revenue.
SemiAnalysis’s GDPx v1.0 scores frontier agents on 11 domains. Opus 5.5 leads at 12%, ahead of Carbon and Astra.
X Post Facto · 25–31 Oct 2026
Google marked Gemini API Computer Use generally available on the Interactions and GenerateContent APIs for GUI agents.
Anthropic tasked Claude agents with writing fine-tuning data that looks harmless yet teaches misalignment past a training-data monitor.
Cursor says tampered Claude Code extension builds 2.1.312 and 2.1.313 sat in its marketplace on Oct. 29 and were not made by Anthropic.
Pydantic’s harness docs describe BoxdSandbox, which runs coder, shell, and file tools in an isolated boxd machine instead of on the host. The page does not document model fallback or a release version.
Claude Code 2.1.316 adds shared group budgets, a subagent effort setting, and an ultrathink off switch, and dates the end of Opus 4.8 fast mode.
OpenCode 2.2.4 keeps desktop terminals across restarts, adds opt-in artifacts, and walks users through connecting a Claude Agent login.
Microsoft Agent Framework Python 1.24.0 lets agents compact context mid-run and warns that Python 3.10 support is ending.
Vercel Labs released agent-desktop, a CLI and C library that lets coding agents operate desktop apps through the OS accessibility tree on macOS, Windows, and Linux without embedding a model.