News

Releases & Models

AI Radar 11/05/2026: Cat Wu at OpenAI, Meta buys Conductor and Claude Code cheapens subagents

Claude Code’s product lead joins OpenAI, Meta acquires Conductor (multi-agent coding orchestration), Claude Code cuts background-subagent status-line cost by about 99%, OpenAI shows a pipeline that found and remediated 948 Linux privilege escalations, and Nous opens a Hermes plugin hackathon with US$20k. We separate what is confirmed from what is still rumor.

Capa do Radar de IA 05/11/2026, estilo radar escuro

How to read this radar

This is the daily roundup of the main AI news circulating on X. We mark as CONFIRMED what came from an official company account or page, as X SUMMARY what only appears in X trend summaries (with no official confirmation we verified), and as RUMOR what has not been confirmed yet.

OpenAI

CONFIRMED: OpenAI launched Images 2.6 with GIFs and reference images in ChatGPT, plus Flare and Sunburst models in the API and Codex, with image generation up to 50% faster. Full story here (OpenAI; OpenAI Devs).

CONFIRMED: Cat Wu, who led Claude Code product at Anthropic, announced she is joining OpenAI to lead product management for Core Products. Codex’s Tibo welcomed her on X (Cat Wu; Tibo). Hours later, Austin Lau, an early Anthropic growth hire, also said he joined OpenAI (Austin Lau).

CONFIRMED: OpenAI described an autonomous cyber vulnerability discovery, exploitation and patching pipeline. In one run with an upcoming cyber model, it validated and generated candidate patches for 948 local privilege-escalation bugs in the Linux kernel, at about US$1.2M in API cost (OpenAI; post). Tibo also flagged the cyber team (post).

Already in yesterday’s radar: the Pierre Computer Company acquisition to boost Codex. Details in the 11/04 radar.

RUMOR: TestingCatalog and X chatter say GPT-6.2 Astra and Sol showed up in Google Cloud quotas and could ship soon. OpenAI has not announced this (TestingCatalog; Decrypt).

Anthropic and Claude Code

CONFIRMED: Claude Code (v2.1.316+) made background-subagent status lines cheaper: instead of re-reading the whole conversation every ~30 s, Haiku 5.5 writes from short notes. Anthropic says about 99% less per refresh and about 5% less of the session limit with background subagents (~6% more prompts). Live for most Pro/Max/Ultra users and everyone on Team/Enterprise; the subagent itself keeps its configured model (ClaudeDevs; follow-up). For anyone running several agents in parallel, this is the kind of tweak that shows up on the monthly bill.

CONFIRMED: Anthropic published how Claude helped map the first 3D structure of a vault particle from bacteria, speeding steps that would have taken weeks (research; Anthropic). The same research hub also notes progress on cryptographic proofs (FlashProve).

Duet: no expansion outside the US and no general availability; still a US-only Pro/Max research-preview waitlist (help center). RUMOR: Fable 5.5 / Mythos 5.5 remain leaks and code demos only, with no announcement (Decrypt).

Meta and Conductor

CONFIRMED: Conductor, a platform for running multiple coding agents (Claude Code, Codex and the like) in parallel, announced it is joining Meta. New sign-ups closed; Conductor Cloud goes read-only on Nov 13 and shuts down on Nov 20 (export data first); the local app keeps models and critical fixes until early 2027, with something new promised early next year (announcement; TechCrunch). On X, open-source alternatives such as omux are already circulating for people who used Conductor with Claude Code or Codex.

Hermes, OpenClaw and SpaceXAI

CONFIRMED: Nous opened the Hermes Agent Plugin Hackathon: 7 days, US$20k in cash prizes (US$10k for 1st), deadline Nov 11 at 5 PM EST. Path is a PR to the plugin catalog, a demo video under 90 seconds on X tagging @NousResearch, and the announcement form (announcement; catalog submission). Accepted plugins land in the Hermes Marketplace (launched yesterday).

CONFIRMED: Tencent’s Hy4 is 80% off on the Nous Portal for one week, with vision and computer use (Nous). Full model write-up here.

Grok 4.8: still no official launch; what ships today is Grok 4.7. The @xai account on X had no useful public posts in this window.

Open opportunities

Hermes Agent Plugin Hackathon: through Nov 11 5 PM EST, open to anyone online, US$20k cash. Fits a Hermes multi-agent fleet directly (announcement).

OpenAI DevDay Exchange São Paulo: tomorrow, Nov 6 at 2 PM; registration still under review (register).

Hy4 80% off on Nous Portal: one week (Nous).

OpenAI/Anthropic/xAI startup credit programs still effectively need VC or a closed cohort: Grok for Startups requires venture backing; Claude Startups’ current cohort is full. Nothing new in this round that fits a non-VC Brazil company.

Also on MaxA: Images 2.6 and yesterday’s radar.

Sources

x.com/_catwu/status/2118137612577968402; x.com/thsottiaux/status/2118138648214585591; x.com/helloitsaustin/status/2118161804463595931; x.com/OpenAI/status/2118124944928432327; x.com/OpenAIDevs/status/2118124753340756055; openai.com/index/a-thousand-fewer-paths-to-root; x.com/OpenAI/status/2118119451401048497; x.com/ClaudeDevs/status/2118063143167693011; anthropic.com/research/investigating-vault-particles-with-claude; x.com/AnthropicAI/status/2118073665296228863; conductor.build/announcement; techcrunch.com/.../meta-acquires-ai-coding-startup-conductor; x.com/NousResearch/status/2118094256438936062; x.com/NousResearch/status/2118136505046904891; hermes-agent.nousresearch.com/.../catalog-submission; events.openai.com/devdayexchange2026; x.com/testingcatalog/status/2118268779901628880; decrypt.co/.../claude-mythos-fable-5-5-gpt-6-2-astra-leaks; support.claude.com/.../claude-duet.

"}