
Anthropic ships Claude Fable 5.2 and retakes the top LLM rank
Artificial Analysis ranks Fable 5.2 first at 65, ahead of Gemini 4 Carbon at 60.
693 stories, newest first · one card per story

Artificial Analysis ranks Fable 5.2 first at 65, ahead of Gemini 4 Carbon at 60.

An Anthropic engineer says each painting is a 1,000-to-2,000-line Python program driving a simulated paint engine, with no external image model.

Amodei told JPMorgan investors Anthropic has committed nearly $901 billion, mostly to compute, with $71 billion due by September 2027.
Testing put Carbon at 85.4% and about $15.70, versus 80.6% and $1.07 for GPT-6.1 Sol.
Anthropic says Medium effort matched predecessor Opus 5 on High, and phrases like "think hard" no longer help.

In a Decoder interview, the former OpenAI director discusses alignment, regulation, China, and the failed nonprofit board.

A November 13 blog post uses a Bill Gates interview to link success bias with AI doom and utopia claims.

Jacob Coxon says an IPO roadshow slide on accelerating efficiency led him to quit after about four months.
A coordinator assigns cloud threads that open pull requests, while shared memory and a file library persist across work.
Whiteboards and Sheets are in beta on paid plans; Claude Data is generally available, with some features still in beta.

Dwarkesh cites METR, Anthropic, and OpenAI figures to argue frontier AI R&D is already accelerating.

The 825 million-share sale could raise as much as $64.35 billion, with Nasdaq trading targeted near November 19.

The Team and Enterprise plugin connects to Autodesk Fusion, NetSuite, Microsoft 365 and Infor but does not control machines.

The research preview is limited to Pro, Max, and Ultra subscribers in the US and links email, calendar, Docs, and Slack.

Dave Ricks said biology is still largely unmapped and human proof, not model forecasts, is the long pole in drug development.

He reports no false positives in that batch and says about 200 bugs were fixed in four days.
The November 13 status note also lists Claude Console and Claude Cowork, with no cause or restoration time.

Customers run the agent server, PostgreSQL session store and workers; Claude model calls still leave the network.

In a lab test, Darktrace exfiltrated a 1 GB file through Anthropic APIs and has not seen the method in the wild.

Among more than 85,000 Ramp customers, a typical firm spent $16.32 while the top 1% spent $10,200 per employee.

Only two of 442 tracked CVEs were exploited, while duplicates and withdrawals far outnumber the 217 marked fixes.

Anonymous sources put the gain near 50% from a reported $6.7 billion, with the margin still negative.
Co-founder Tom Brown says Claude is trained and served on AWS Trainium, Google TPUs, NVIDIA GPUs, and AMD GPUs.

Adjusted operating losses still rose to $10.9 billion, while Anthropic reported $17 billion in revenue.
Today 1.2K readers · 1.5K views · 295 reads · all time 200K views