How to read this radar
This is the daily roundup of the main AI news that circulated on X. We label as CONFIRMED what came from an official company account or page, as X SUMMARY what only shows up in X trend summaries (with no official confirmation we could verify), and as RUMOR what has not been confirmed yet. Items tagged "5 p.m. update" were added in the afternoon edition.
OpenAI and Codex
5 p.m. update CONFIRMED: Tibo said OpenAI's dots agents will be powered by more frontier models, including GPT-6 Astra and GPT-6.1 Terra, and that each dot will pick the best model for each part of the work (post; detail). According to him, Terra brings speed and a nicer personality and will help bring dots to more people.
5 p.m. update CONFIRMED: OpenAI introduced LifeVal, a benchmark of everyday tasks built from the feedback of thousands of ChatGPT users, from finding childcare to accessing government services and managing personal finances, which it has used internally for a year (OpenAI; official page).
5 p.m. update CONFIRMED: at ADIPEC in Abu Dhabi, Sam Altman said there is no room for debate about keeping humans in control of AI, and ADNOC and G42 signed a deal with OpenAI to use its models across the oil company's operations (The National; Abu Dhabi government). He also retold the story of 10,000 agents spending 88 hours on the Navier-Stokes proof with an unreleased GPT; that proof is still disputed by mathematicians.
5 p.m. update RUMOR: the ID "gpt-6.2-astra" has been showing up in some users' Codex usage analytics since Oct 31 while the picker still says GPT-6 Astra, which suggests an A/B test or staged rollout (TestingCatalog). OpenAI has announced nothing. The same site claims GPT-6.1 Astra was shelved in September, which is not officially confirmed either.
CONFIRMED: the 28-day Codex challenge is over. Tibo (thsottiaux) closed it with "Day 28/ Banked 2/2", a second reset saved to accounts (post), after the first one (Banked 1/2). OpenAI explained how these resets work: you pick when to use them in Usage on desktop or web, or with /usage in the CLI; mobile can't redeem, and a reset only applies when some allowance is below 80% (help center).
CONFIRMED: on day 26 Codex got switching between multiple accounts on mobile, with desktop coming soon (post).
CONFIRMED: the Agents API gained session-level auto review (review: "auto_review", off by default) and async compaction on by default, which may use more tokens (OpenAIDevs; docs).
RUMOR: the Arena codename "fe-raichu" would be GPT-6.1 Astra (post), and "expected in November" model lists are circulating again (post). Nothing official.
Anthropic
5 p.m. update CONFIRMED: Anthropic filed its first S-1 amendment with the SEC today, adding the third quarter: $16.96 billion in quarterly revenue and $33.24 billion over nine months, a $23.29 billion net loss for the quarter (mostly fair value changes on convertible notes) and about $895.4 billion in commitments, nearly all compute. The filing also says future models may soon hit the "dramatically accelerated progress" threshold in its Responsible Scaling Policy (SEC; CNBC). The IPO itself was covered in the Oct 27 radar.
5 p.m. update CONFIRMED: applications opened for the Anthropic Residency, a paid program lasting six weeks to six months for quantitative professors, technical founders and staff-level engineers who want to move into AI safety research, with no ML background required; it closes Nov 22 (AnthropicAI; official page).
5 p.m. update RUMOR: demos of HTML-only animations attributed to a "Claude Fable 5.5" are circulating (trend). The latest official Fable is still 5.1.
CONFIRMED: starting November 3, Claude's text watermark also covers Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6 worldwide, tied to Article 50 of the EU AI Act (help center; Mashable).
CONFIRMED: Anthropic published research using thousands of Claude agents to automate red-teaming for subtle emergent misalignment (AnthropicAI).
X SUMMARY: the Messages API was reportedly rewritten in Rust with Claude's help, using 40% to 50% less CPU (trend).
RUMOR: a hidden "Claude Mythos 5.5" reference reportedly showed up in Anthropic's site HTML (trend). Fable 5.5 is still unannounced.
Duet: no expansion. It is still a US-only waitlist for Pro, Max and Ultra, with Max invites first and international "no date" (help center). People admitted get one full limit reset, valid until Nov 30. We covered the launch here.
SpaceXAI and Grok
5 p.m. update CONFIRMED: Grok Bot users saw their weekly limit hit 100% over the weekend even with idle bots, and Elon Musk replied that "the cost of serving Grok Bot was unsustainably high relative to revenue" (post; TechCrunch; Business Insider). No new limit numbers were published: the official page still lists Premium at $30, Super at $100 and Ultra at $200, and Grok Bot usage comes out of one weekly pool shared with Grok and Cursor (pricing; docs). If you run several bots, check the meter before leaving heavy routines scheduled.
CONFIRMED (docs, via press): from November 7 the Grok API turns on text watermarking by default for grok-4.7, 4.6, 4.5, 4.3 and the 4.20 family (including multi-agent). Admins can opt out per model when a legal exception applies, and responses carry an x-watermark: true header (TechCity Authority).
X SUMMARY: Grok Bot reportedly passed 600k daily active users (trend) and people are building agent teams on it (trend). Grok 4.8: nothing new beyond rumor lists.
Agents: Hermes, OpenClaw and Cursor
5 p.m. update CONFIRMED: Hermes gained the hermes migrate grok-bot command (plus openbot), and Nous teased it by quoting Musk's post (NousResearch; PR). Since Grok Bot has no export button, each bot writes its own grok-bot-export.json from a message Hermes provides. Scheduled routines become paused cron jobs with the same schedule and time zone, skills and memories come along, the bot's role goes to SOUL.md and remote MCPs land in the config; event-triggered routines become a .md file to rebuild by hand, and credentials are not copied. You can run it with --dry-run first (docs).
5 p.m. update CONFIRMED: OpenClaw opened bi-weekly OpenClaw Enterprise office hours for people building agents at work to bring questions and help shape the open, vendor-neutral platform (post; registration).
5 p.m. update X SUMMARY: Open Duet, an open-source, self-hosted project for "AI coworkers" with roles, memory, routines, a lead agent that delegates to specialists and approval before acting, has reportedly passed 6,000 GitHub stars (repository; trend). The repository exists; the star count and quality were not tested by us.
CONFIRMED: Nous Research opened the Hermes Mobile beta, an app to control your own Hermes from your phone (official page). iOS is via TestFlight; your Hermes API server/gateway must be reachable (Tailscale works). Testers reported Android seats full on Oct 31, and fake Hermes apps are already in the stores, so install only from the official link (first look; NousCon recap).
CONFIRMED: Cursor and SpaceXAI opened an ambassador program with free Cursor credits, nightly early access, meetup funding, a private Slack and merch (Ben Lang; page).
X SUMMARY: "harness engineering" is the topic of the moment among agent builders (trend).
Voice
X SUMMARY: VoiceArena's Turing TTS benchmark measures a "fool rate", and Gemini 3.8 Flash TTS leads at 10.5% (trend).
X SUMMARY: the OpenCode team released Hex, local open-source dictation (trend), and YC-backed Subvocal glasses opened preorders at $499 (trend).
Hermes Mobile also brings a live voice mode, according to testers. In the afternoon edition we found no new confirmed voice news.
Open opportunities
5 p.m. update Anthropic Residency: for quantitative professors, technical founders and staff engineers, no ML required; pays the equivalent of $350k to $500k a year, pro-rated; remote is possible for shorter placements as long as you are in a country where Anthropic can hire. Deadline: Nov 22, 11:59 p.m. Pacific (apply).
5 p.m. update OpenClaw Enterprise office hours: free and bi-weekly, register at this link.
Cursor × SpaceXAI ambassador: 18+, no fluent English required, weekly review with an answer in 2 to 3 weeks and no deadline. Apply via this form (the old button on the page points to a closed form).
Hermes Mobile beta: iOS via TestFlight, while seats last.
Codex resets: if you banked resets during the challenge, redeem them before the expiry shown in your account (how it works).
The Gemini 4 Carbon Arena rumor is already in this story, and the previous radar is here.
Sources
x.com/thsottiaux/status/2117035794678546764; help.openai.com/en/articles/20001498-how-resets-work-in-codex; x.com/OpenAIDevs/status/2116306164375863343; developers.openai.com/api/reference/resources/beta/subresources/agents/subresources/sessions; x.com/thsottiaux/status/2117283124833055196; x.com/OpenAI/status/2117307611338997846; openai.com/index/lifeval; www.thenationalnews.com/future/technology/2026/11/02/no-room-for-debate-over-human-control-of-ai-says-sam-altman; www.mediaoffice.abudhabi/en/crown-prince-news/khaled-bin-mohamed-bin-zayed-meets-sam-altman-and-witnesses-signing-of-agreements-with-openai; www.testingcatalog.com/gpt-6-2-astra-appears-codex-usage; x.com/AnthropicAI/status/2116349211549843666; support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content; sea.mashable.com/tech/56556/anthropic-expands-text-watermarking-to-older-claude-versions-beginning-november-3; support.claude.com/en/articles/17631495-get-started-with-claude-duet; www.sec.gov/Archives/edgar/data/2130357/000162828026068676/anthropicpbc-sx1ax1.htm; www.cnbc.com/2026/11/02/anthropic-ant-s-1-quarter-financials.html; x.com/AnthropicAI/status/2117343929960112486; alignment.anthropic.com/2026/anthropic-residency; x.com/elonmusk/status/2116992258520064397; techcrunch.com/2026/11/02/elon-musk-throttles-grok-bot-limits-due-to-unsustainably-high-costs; www.businessinsider.com/grok-bot-usage-limits-costs-unsustainably-high-elon-musk-agents-2026-11; x.ai/pricing; docs.x.ai/x-subscription/grok-bot; www.techcityauthority.com/2026/11/spacexsi-grok-api-text-watermarking-default-november-7.html; x.com/NousResearch/status/2117326444669857941; hermes-agent.nousresearch.com/docs/user-guide/import-from-other-agents; github.com/NousResearch/hermes-agent/pull/147724; hermes-agent.nousresearch.com/mobile; testflight.apple.com/join/CVhbq8Ur; x.com/openclaw/status/2117300313078337812; openclaw.ai/enterprise#webinar; github.com/Anil-matcha/open-duet; x.com/benln/status/2115499001864700343; anysphere.typeform.com/spacexsi-amb; cursor.com/ambassadors; x.com/Smlay_ero/status/2116456722328166759; x.com/i/trending/2116967041924010487; x.com/i/trending/2117172111106355474; x.com/i/trending/2116630671536066855; x.com/i/trending/2117293628175323619; x.com/i/trending/2116319406649122984; x.com/i/trending/2116678391482953816; x.com/i/trending/2117023502666469859