Anthropic Launches Claude Mythos 5.2 and Fable 5.2: Anthropic released Claude Mythos 5.2 and Fable 5.2, reducing API token pricing by 20% to $8 per million input and $40 per million output tokens while speeding generation by up to 30% . The update delivers significant latency and cost improvements for long-horizon autonomous agents, with Anthropic reporting the models resolved benchmark challenges in higher mathematics and black hole physics.
Google Unveils TPU 8tx and TPU 8i Max for Cloud AI: Google announced two specialized eighth-generation tensor processing units arriving in 2027: the TPU 8tx, engineered with AMD EPYC CPUs for confidential workloads, and the dual-chip TPU 8i Max, doubling compute density and memory bandwidth for long-context inference The rollout marks Google’s first release of four distinct TPU configurations in a single year, intensifying competition with Nvidia by targeting specialized enterprise inference and security needs.
Federal Inquiries Expand Around Nvidia Over Foreign Espionage Concerns: The US Department of Justice assumed control of an investigation into GPU distributor Megaspeed over alleged chip smuggling to China, while President Trump confirmed the FBI is reviewing whether foreign intelligence targeted CEO Jensen Huang . The escalating federal probes add significant regulatory and export-control headwinds to the leading supplier of global AI hardware.
Meta Debuts Muse Ember for High-Volume Agent Work: Meta launched Muse Ember, a lightweight reasoning model priced at $0.10 per million input tokens designed specifically for high-throughput, routine agent tasks like tool execution, classification, and code repairs . The release reflects an industry-wide pivot toward compact execution models that prevent repetitive autonomous operations from consuming expensive frontier compute budgets.
Disney and Adobe Partner on Custom Generative AI Production: The Walt Disney Company named Adobe its Official AI Innovation Partner to build proprietary generative AI models using Adobe Firefly Foundry trained exclusively on Disney’s intellectual property for parks, ESPN, and StudioLAB . The multi-year alliance establishes a prominent commercial model for creative studios adopting licensed, commercially protected generative tooling for production workflows.
SpaceX Plans $24B Private Supercomputer Campus at Starbase: SpaceX submitted plans to build a dedicated natural-gas power plant and supercomputing campus in Starbase, Texas, representing approximately $24 billion in initial infrastructure investments . The development demonstrates how major AI and space technology ventures are bypassing strained regional electrical grids to construct self-reliant, behind-the-meter generation for high-density clusters.
Fireworks AI Discloses Data Breach Compromising Model Weights: Cloud AI infrastructure provider Fireworks disclosed that unauthorized attackers accessed select customer datasets, billing records, and partial model weights over the preceding month . The breach highlights emerging supply-chain security risks as third-party model hosting and fine-tuning platforms become primary targets for intellectual property theft.
