Quick answer
Claude Sonnet 6 and Fable 6 names are still rumours — but METR’s Anthropic assessment and GPT-6 Cyber are the week’s firmer stories. WorldofAI covers leaked Claude checkpoints, a METR report claiming Anthropic’s AI work is progressing about 1.9× faster, OpenAI’s security-focused GPT-6 Cyber, Gemini Omni 1.2 Flash, Meta stealth models, Claude’s bank-connect beta, and politics around data centres. For a small business, treat unreleased model names as speculation, read METR carefully, and only act on shipped features.
Key takeaways
- Sonnet 6 / Fable 6 labels are unconfirmed. WorldofAI shows strong SVG demos from a rumoured newer Claude checkpoint — not an official release.
- METR says Anthropic’s progress may be ~1.9× faster, with AI helping drive much of that acceleration, and describes three unreleased internal models.
- GPT-6 Cyber launched for select security customers via OpenAI’s Daybreak Red programme; Astra/Sol remain reported as imminent.
- Claude Personal Finance (US beta) can connect banks read-only via Plaid, per the video.
- Google Vids / Omni 1.2 Flash and several open or small models also shipped or demoed.
- Your action: verify leaks before switching tools, and decide deliberately before linking a bank account to any AI.
In this guide
- What are the Claude Sonnet 6 / checkpoint rumours?
- What did the METR report say about Anthropic?
- What is GPT-6 Cyber and is Astra next?
- What Meta models are being stealth-tested?
- What is Claude Personal Finance?
- What did Google ship with Gemini Omni 1.2 Flash?
- Did midterms and data-centre politics change AI?
- What other tools and models made the list?
- What should a small business do this week?
- What does the video not confirm?
- Watch: the source video
- Frequently asked questions
- The bottom line
- Related posts
AI news weeks that mix leaks, papers and product launches are easy to misread. WorldofAI’s round-up is useful if you separate what shipped from what is still a rumour.
Credit for the round-up goes to WorldofAI (@intheworldofai); the judgements here are ours. The video is sponsored by Soulog Sense.
What are the Claude Sonnet 6 / checkpoint rumours?
WorldofAI opens with speculation that Anthropic may be routing some Sonnet 5.5 “extra high” traffic to a newer unreleased checkpoint — sometimes called Sonnet 5.6 or Sonnet 6 in community chatter. It is not presented as confirmed Fable 5.5.
The standout demo in the video: a code-only pelican-on-a-bicycle SVG with tens of thousands of elements. Treat that as a capability teaser from an unknown checkpoint, not a product you can buy today. For leak hygiene, see our guide on verifying AI model leak rumours.
🔎 Unconfirmed / rumour: Sonnet 6, Opus 6, Fable 6 and Mythos naming for METR’s “Model 2/3/4” are community speculation. Anthropic has not confirmed those product names in the video.
🛠️ Practical guides
Ready to put this into practice? These step-by-step guides answer two common follow-up questions:
- How to Read a METR AI Acceleration Report Without Overreacting — separate METR evidence, uncertainty and rumour before you change tools
- Should You Connect Your Bank Account to Claude? — decide whether Claude Personal Finance is worth the privacy trade-off
What did the METR report say about Anthropic?
METR published an independent assessment of Anthropic’s internal AI R&D. WorldofAI highlights:
- Progress roughly 1.9× Anthropic’s previous rate, with AI possibly responsible for around 80% of the acceleration (as summarised in the video).
- Access to three unreleased models (Model 2, 3, 4) built between July and October 2026.
- Model 4 reportedly scoring above 50% across Anthropic’s 15-benchmark capability index, with a score around 178.6 vs Opus 5.5 around 174–175.
- Employee self-reports of about 12× productivity with the best internal model (earlier survey ~4×) — self-reported and biased.
- No convincing evidence of major compute/data/algorithm bottlenecks in the next 12 months, with high uncertainty.
- Anthropic redacted material; co-founder Sam McCandlish reportedly said that was to avoid handing competitors a roadmap.
Our guide on reading a METR acceleration report without overreacting shows how to use these numbers without rewriting your stack overnight. We also covered earlier RSI framing in Anthropic recursive self-improvement.
What is GPT-6 Cyber and is Astra next?
OpenAI officially introduced GPT-6 Cyber for advanced cyber-security research, initially for select Daybreak Red customers. WorldofAI cites ExploitGym scores of about 48.5% for Cyber vs 42.4% for Astra. GPT-6.2 Astra and Sol remain “as early as Friday” style reports, with a Codex catalog sighting and a playful OpenAI reply — still not a launch.
What Meta models are being stealth-tested?
Meta’s Wonderful line allegedly has an internal “restful-walrus” checkpoint. Stealth models Crouton Alpha and Pulse Alpha appeared on OpenCode/OpenRouter-style tests: fast, multimodal, with Crouton described as roughly GPT-6.1 Terra-speed but weaker quality. Interesting for experimenters; not a reason to migrate production.
What is Claude Personal Finance?
Claude’s Personal Finance beta (US) connects bank accounts through Plaid for natural-language spending questions. WorldofAI and Matt Wolfe both covered the read-only framing. Our guide Should you connect your bank account to Claude? walks through a decision checklist.
What did Google ship with Gemini Omni 1.2 Flash?
Gemini Omni 1.2 Flash: longer clips (up to 15s), lower price, better reference adherence, available in API/Flow/Vids, with free monthly clips in Vids. This overlaps Matt Wolfe’s round-up; see also our Google Vids guide.
Did midterms and data-centre politics change AI?
Post-midterm politics: pressure on data centres, Texas governor-elect comments, reports OpenAI may revisit 2027 capacity plans, and a Google $17 billion Wyoming investment cited in the video. Markets sold off hard in AI names while Google still commits capital. Treat political forecasts as opinion; use our AI rules checklist for what actually applies where you trade.
Hearing leak fatigue? Use our METR reading guide before you change tools, and keep a hard line between rumours and release notes.
What other tools and models made the list?
| Item | What the video says | Action |
|---|---|---|
| Cursor app deploy | Build/deploy/share apps inside Cursor with security checks | Try if you already live in Cursor |
| Inkling 2 Small | Smaller model with stronger index score; weights planned | Watch for local fine-tunes |
| Ling 3.2-flash | 560B MoE, 25B active; multimodal claims | Benchmark for your tasks first |
| Agent Desktop | Agents control the desktop with separate cursors | Sandbox before any real accounts |
| Mouse surgery update | In-context learning across three procedures | Interesting research; not a product |
| Muse realtime avatar | Talking avatars from images; Meta evals vs rivals | Consumer novelty; check rights |
What should a small business do this week?
- Do not switch Claude tiers based on Sonnet 6 rumours.
- Read METR (or a careful summary) for orientation, not panic.
- Ignore GPT-6 Cyber unless you are in a security programme.
- Decide on Claude bank connect with a written privacy check.
- Try one shipped creative feature (Omni clip or Images GIF) on a low-risk asset.
What does the video not confirm?
🔎 Unconfirmed / rumour: Claude Sonnet 6 / Fable 6 naming, METR productivity multiples, Astra/Sol timing, Meta code-names, stock-loss figures, “40% of OpenAI capacity” claims and the mouse-surgery results are as stated in the video or secondary reports. Flag them as unconfirmed until primary sources match.
Watch: the source video
What this video covers: WorldofAI’s AI news: Claude checkpoint and Sonnet 6 rumours, METR’s Anthropic acceleration assessment, GPT-6 Cyber, Astra speculation, Meta stealth models, Claude Personal Finance, Gemini Omni 1.2 Flash, midterm/data-centre politics, Cursor deploy, small models, Agent Desktop and a surgery research update. Sponsored by Soulog Sense.
Video: WorldofAI (@intheworldofai). Sponsored by Soulog Sense. This summary and the opinions in it are our own. Leaks, METR figures, stock moves and unreleased model names are rumours or secondary reports until the companies confirm them.
Frequently asked questions
Is Claude Sonnet 6 out?
Not as an official public release in this video. Treat names and demos as rumours until Anthropic announces.
Should I trust the METR 1.9× figure?
It is an independent assessment summarised by the creator, with redactions and uncertainty. Use it as a signal, not a schedule.
Can I use GPT-6 Cyber?
The video says access starts with select security customers through Daybreak Red.
Is linking a bank to Claude safe?
Anthropic describes read-only Plaid access and no training on the data. You still share sensitive information — decide carefully with our checklist.
Is the AI boom ending after the midterms?
The creator argues no, citing continued investment such as Google’s Wyoming plan. Politics can still slow local data centres.
The bottom line
This week’s noise is mostly leaks and politics; the usable edges are clearer: a security model for a narrow audience, creative video/image tools, and a finance beta that needs a hard privacy choice. Keep rumours in the rumour box, and only change workflows when a lab ships release notes.
Put it into practice: read METR without overreacting, then decide whether Claude should see your bank data.
Related posts
- Claude Mythos 5.5 Leak: Fable 5.5 Demos, GPT-6.2 Astra and This Week’s AI News
- Calls to Pause AI: What the Midterm Debate Means, Plus GPT-6.2 Sol and This Week’s AI News
- How to Verify AI Model Leak Rumours Before You Switch Tools
