How to read this radar
A roundup of what circulated on X and official sources since yesterday afternoon. CONFIRMED means it came from an official account or page; X SUMMARY means it only shows up in X trend summaries; RUMOR means it is still unconfirmed.
Grok and SpaceXAI
CONFIRMED: Musk said Grok 4.8 pre-training is done, the model is in RL with evals climbing, and there will be a soft rollout to Grok Bot users later this week. xAI dropped JAX and SGLang and rewrote its whole RL and inference stack in C/C++, which explains the delay. He says 4.8 is "not yet incredible" but "will bear no resemblance to 4.7" (Musk). No exact date or rollout scope yet.
5 PM UPDATE, CONFIRMED: asked how Grok 4.8 compares with Claude Opus 5.5, Musk replied "not as good", called Opus 5.5 "a masterpiece" and said Grok 5 should exceed it "maybe in 2 or 3 months" (Musk). The official Grok Bot account reset weekly limits for all users yesterday (Grok Bot). X SUMMARY: this week's Grok Bot update brings phone calls placed by the bot (with a live transcript and approval first, for now US only, excluding Washington and Illinois), a share sheet, Team Bots, warnings for hidden instructions on web pages and a usage screen by bot and by source (what is new video; X summary; X summary).
Anthropic and Claude
CONFIRMED: Anthropic launched the roadshow for its proposed IPO; the SEC registration is not yet effective (Anthropic; post).
CONFIRMED: Anthropic showed what Claude Fable 5.2 did in testing: from a one-paragraph prompt it designed three working electronic devices, including a home voice assistant hub (Claude). X SUMMARY: in another demo the model coordinated 1,000 subagents over eight days to build a bootable operating system (X summary). API users: Becca already explained why just swapping the model ID returns a 400 error: Fable 5.2 returns error 400 if you only swap the model ID.
X SUMMARY: Fable resets only restore the Fable-specific share of usage and don't help once the weekly cap is maxed; they expire around March 9, 2027 (X summary). In T3 Code, Opus 5.5 now handles 75% of user turns (X summary).
X SUMMARY: on 60 Minutes, Dario Amodei backed U.S. military use of Claude to speed up targeting (X summary). There's also heated debate over Anthropic capping Fable's training on kernel optimization and LLM architectures. RUMOR: unverified reports say it backs firmware locks on Western hardware against Chinese models (X summary). RUMOR: Fable 5.5 is still unannounced.
5 PM UPDATE, CONFIRMED: Anthropic started rolling out Claude Whiteboards (boards to plan), Claude Sheets (spreadsheets) and Claude Data (data storytelling) today (Claude). X SUMMARY: they are betas for paid plans on web, desktop and mobile, aimed at Miro, Excel and Tableau (X summary).
X SUMMARY: Fable 5.2 scored 41.3% on the Remote Labor Index, a test built from 240 real Upwork projects, more than double Fable 5.1's 17.9% (X summary). On the IPO roadshow, the cited range is $70 to $78 per share, which would value Anthropic at up to about $1.4 trillion (X summary). Claude Duet: the team is asking preview users for suggestions and users are asking for EU access, but no expansion was announced; it remains a US preview (X summary).
OpenAI
CONFIRMED: OpenAI unveiled Control Room, a central system that tracks every model workload on its research infrastructure, including agentic sampling, and can halt model output within 35 seconds of a decision to intervene (OpenAI; post).
X SUMMARY / RUMOR: some ChatGPT Plus users report early access to what looks like GPT-6.2 Sol, with strong 3D and coding results; OpenAI hasn't announced it (X summary). GPT-6.1 Astra is still unreleased.
5 PM UPDATE, CONFIRMED: OpenAI put Businesses in ChatGPT in beta, letting companies claim and manage their listing (OpenAI); Becca covered it in full here: OpenAI opens business listings in ChatGPT, but agents start in the US. OpenAI also scheduled Cyber at the Frontier, a live keynote with its latest cyber announcements, for Thursday, November 12 (OpenAI).
Hermes, OpenClaw and agents
CONFIRMED: Claude Fable 5.2 is now usable in Hermes Agent (Nous).
X SUMMARY: Ethan Mollick let 25,000 agents build 3D worlds; they self-organized into a hierarchy with leads, theme teams and auditors, made 450+ worlds and wrote an 87,500-word rulebook. Output fell from 211 worlds on day one to two a day by day ten, slowed by bureaucracy (X summary). A useful lesson for anyone orchestrating agent fleets.
5 PM UPDATE, CONFIRMED: Sign in with ChatGPT now accepts the ChatGPT Business plan in tools like Notion, v0, Amp, Kilo Code, Warp, Hermes, OpenClaw and OpenCode (OpenAI Devs), and Nous Portal confirmed support (Nous). Teams running agent fleets can now hang Hermes and OpenClaw on their Business subscription. CONFIRMED: Gemini 4 Carbon is now in Cursor and, at high effort, tops CursorBench at 60.4% (Cursor).
Voice
X SUMMARY: AirCaps, a five-person startup, launched A5S Async V1, which tops the speaker diarization benchmark ahead of ElevenLabs, AssemblyAI and Deepgram, processing an hour of audio in 9.8 seconds for $0.18 via API (X summary). Relevant for agents that join meetings.
5 PM UPDATE, X SUMMARY: Fish Audio launched Live 3, a full-duplex voice model that listens while it talks and reacts in under 200 milliseconds to interruptions and "mm-hmm"; it runs on the Fish Agent platform for phone and web, English first (X summary). Grok Bot's phone calls (see Grok section) also count as voice agent news.
Open opportunities
Grok 4.8: the initial rollout targets Grok Bot users, per Musk (post).
Voice credits: the TheGrantIndex aggregator lists as open worldwide the programs from Telnyx ($20K for voice), Deepgram (up to $100K), Speechmatics (up to $50K) and Smallest.ai ($10K). Check each official page before applying (TheGrantIndex).
5 PM UPDATE, Zoom AI Launchpad: Zoom opened a credits program today for speech, language and conversational AI startups, equity-free: $5K, $50K or $150K in Zoom AI Services credits for 12 months, reviewed on a rolling basis with limited spots (press release). Its FAQ says startups worldwide may apply and funding stage is not a strict requirement (apply).
Sources
x.com/elonmusk; www.anthropic.com/news/roadshow-launch; x.com/AnthropicAI; x.com/claudeai; openai.com/index/control-room; x.com/OpenAI; x.com/NousResearch; thegrantindex.com/category/ai-credits/; maxassistant.app/en/news/tools/fable-5-2-four-changes-break-existing-integrations; x.com/elonmusk (Opus 5.5); x.com/bot; x.com/mattyp; x.com/claudeai (Whiteboards); x.com/OpenAI (Businesses); x.com/OpenAI (Cyber); x.com/OpenAIDevs; x.com/NousResearch (Business); x.com/cursor_ai; globenewswire.com (Zoom); zoom.ai/ai-launchpad; maxassistant.app/en/news/releases/openai-businesses-chatgpt-business-listings-agents.
