Claude Fable 5.1 Launch, World Labs Atlas, and OpenAI Legal Updates
Anthropic has released Claude Fable 5.1 with significant cost reductions and improved agentic performance. World Labs unveiled its 3D world model, Atlas. Meanwhile, OpenAI has restarted training for its Astra model, and Apple has filed new evidence in its ongoing legal battle against OpenAI.
Podcast В· 2 min
Anthropic Launches Claude Fable 5.1
Anthropic has released Claude Fable 5.1, an upgrade to its flagship model designed to address user feedback regarding cost and safety friction. The new version features a 75% reduction in cache read costs, now priced at $0.25 per million tokens, and demonstrates significant improvements in long-running agentic tasks, such as coding and scientific research. Additionally, the model's safety filters have been adjusted, with Anthropic reporting an 85% reduction in safety interventions for biology-related queries and a 60% reduction for cybersecurity tasks. This release marks a shift in the frontier model landscape, as Anthropic aims to improve the utility of its models for complex, multi-step agentic workflows. The company also introduced Mythos 5.1, a version with fewer guardrails, currently restricted to screened cybersecurity and biology researchers. Early benchmarks indicate that Fable 5.1 outperforms its predecessor in agentic science tests, scoring 52.6% compared to Fable 5's 24.7%.
World Labs Introduces Atlas 3D World Model
World Labs, the company led by AI researcher Dr. Fei-Fei Li, has introduced Atlas, a new world model capable of generating 3D environments from a limited set of phone video clips. The model analyzes 2D footage to reconstruct a 3D version of the space, allowing for the generation of new camera angles and perspectives that were not originally filmed. Atlas aims to create reusable digital representations of physical scenes, which could have significant applications in robotics training, 3D geometry generation, and content creation. By enabling users to capture a space once and manipulate it virtually, the technology represents a step forward in how AI models interpret and render spatial data from real-world inputs.
Apple Files New Evidence in OpenAI Lawsuit
Apple has submitted new evidence in its ongoing legal dispute with OpenAI, alleging that a former engineer, Chang Liu, utilized stolen Apple designs during his tenure at OpenAI. The filing includes claims that Liu downloaded confidential power-converter circuit designs via cloud storage after leaving Apple and attempted to wipe evidence from his devices following the start of the legal investigation. OpenAI has countered these claims by criticizing Apple's offboarding processes, suggesting that the company's internal handling of departing employees contributes to such issues. This development adds a new layer to the legal conflict, as Apple seeks to address what it characterizes as a hardware-related setback caused by the alleged intellectual property theft.
OpenAI Restarts Astra Training
OpenAI has officially restarted the training run for its upcoming Astra model. The training process had been previously paused following a security breach involving Hugging Face. OpenAI has designated Astra as its first 'Critical' cyber risk, indicating a heightened focus on security protocols during the development phase. While the company has not provided a specific release date, it confirmed that a limited release is expected soon. This restart is a critical step for OpenAI as it looks to compete with other frontier models, with the company emphasizing that future models are being paced to ensure a better understanding of their potential consequences.
Bernie Sanders Calls for Global AI Pause
U.S. Senator Bernie Sanders has published an op-ed in Fox News advocating for an international pause on the development of more powerful AI models. Sanders argues that the technology is advancing beyond the control of the companies building it, citing concerns over energy consumption, environmental impact, and potential job displacement in the tens of millions. Sanders proposed that the U.S. and China should negotiate a 'Cold War'-style agreement to manage AI development. His call for a pause comes amid a series of recent security incidents in the industry, which he uses to highlight the risks of rapid, unchecked deployment of frontier AI systems.