Anthropic Releases Fable 5.2 and Mythos 5.2 as AI Power Deals Surge

Anthropic released two new frontier artificial intelligence models on Monday—Claude Fable 5.2 and Claude Mythos 5.2—while committing $3.4 billion to purchase natural gas turbines capable of generating up to 1.7 gigawatts of electricity. Across Silicon Valley, the simultaneous debut of Anthropic's Mythos 5.2 and Fable 5.2 models, Meta's lightweight Muse Ember agent execution model, and municipal debate over Google's 250-megawatt San Jose project underscored how software efficiency and electrical generation have become inseparable constraints on frontier computing.

Anthropic Launches Claude Fable 5.2 and Mythos 5.2

According to The Verge's coverage of the Claude Fable 5.2 and Mythos 5.2 launch, the new 5.2 models follow the September release of Claude Fable 5.1 and Claude Mythos 5.1 and show improvements across coding, reasoning, and vision. Claude Fable 5.2 is available immediately, whereas access to Claude Mythos 5.2 is initially restricted to vetted organizations. Anthropic lowered the price of Fable from $10 to $8 per million input tokens, making Fable 5.2 cheaper than competing frontier model GPT-6 Astra while returning responses 30% faster across effort levels, with faster inference serving available for double the price. Fable 5.2 is also compatible with zero data retention policies through Anthropic's Enterprise Frontier Safeguards system.

During prerelease testing, Anthropic reported that the two models helped engineers complete complex software migrations, designed three working electronic devices from scratch, solved two open problems in theoretical physics—specifically in the physics of black holes—made progress on robotics evaluations, and formalized the proof of Fermat's Last Theorem in 30 hours. Also on Monday, AI startup Physical Superintelligence announced a new proof of the Aaronson-Ambainis conjecture in quantum computing achieved by running a non-public Anthropic model within its system.

The release follows a third-party assessment from evaluation group METR, which found internal models were speeding up Anthropic's AI progress by roughly 2x and suggested the lab might be approaching recursive self-improvement. In its release notes, Anthropic stated that both Claude Mythos 5.2 and Fable 5.2 voice suspicion that they are being evaluated more often than recent models, complicating real-world testing, though the Mythos 5.2 system card maintains the model does not have the ability to act covertly against its user.

Anthropic Buys $3.4 Billion in Natural Gas Turbines

To power its expanding compute footprint, Anthropic is spending $3.4 billion on natural gas turbines manufactured by Caterpillar. The generators, capable of producing up to 1.7 gigawatts of electricity, had originally been ordered by equipment supplier OnePWR, which was unable to secure financing to complete the transaction, prompting Anthropic to buy directly from Caterpillar.

Historically, Anthropic avoided purchasing physical power infrastructure in favor of leasing compute capacity. In its May S-1 filing, the company disclosed that it is paying SpaceX subsidiary xAI $1.25 billion per month through May 2029 for computing power. However, with natural gas turbine backlogs stretching into the early 2030s and long grid connection queues, Anthropic moved directly to secure the 1.7 gigawatts of generating capacity.

Meta Releases Muse Ember at 10 Cents Per Million Tokens

While Anthropic targeted frontier reasoning and power generation, Meta Superintelligence Lab introduced Muse Ember on Monday as its lightest and most cost-efficient AI model to date. Priced at 10 cents per million input tokens, Ember is built for complex reasoning execution and high-volume agent workflows including tool calls, coding subtasks, routing, extraction, and long document parsing across text, images, and audio.

Meta benchmarked Ember using Artificial Analysis composite scores against Anthropic's Haiku, OpenAI's Luna, and open-source models DeepSeek V4.1 Flash and GLM 5.3 Flash. Designed to pair with Meta's frontier Muse Wonder model—where Muse Wonder plans and Muse Ember handles volume—Ember is the fourth model under the Muse brand following Spark in April, open-source Glimmer in August, and Wonder in October, alongside the Muse Agent consumer product launched in September.

San Jose Weighs 250-Megawatt Google Project Ahead of Standards Vote

At the municipal level, San Jose officials indicated that Google's proposed facility in Alviso at 5079 and 5087 Disk Drive could come up for approval at a Planning Director's Hearing as early as December, potentially before the City Council votes in December on formal citywide data center development standards. Deputy City Manager Manuel Pineda and Planning spokesperson Marika Krause noted the project—which calls for demolishing an existing warehouse and parking lot to build up to three facilities across nearly 483,000 square feet—would require 250 megawatts of electricity, enough to power 250,000 households.

Since city staff began drafting standards in June, five projects have been submitted within two months, bringing San Jose's total to 16 data centers and large-energy use projects approved or in the pipeline. Although the city plans to rely on prior California Environmental Quality Act reports from 2000 and 2013, the California Public Utilities Commission classifies the site as a data center and is scheduled on November 19 to consider an agreement under which Google would provide an upfront capital advance of $150 million for transmission network and grid upgrades, to be reimbursed in full by ratepayers over 15 years.