Models & Labs
Unconventional AI Aims to Slash AI Power Use by 1,000x
TechCrunch AIhigh confidence
Why it matters
- →Unconventional AI's architecture could drastically reduce AI's energy consumption, addressing a major scalability issue.
- →The oscillator-based approach offers a novel alternative to traditional computing architectures, potentially revolutionizing AI hardware.
- →This innovation could enable more sustainable AI development, crucial as demand for AI capabilities grows.
Unconventional AI, founded by ex-Databricks AI head Naveen Rao, is developing a novel computing architecture that promises to cut AI power consumption by 1,000 times. The company's first model, Un-0, showcases this technology's potential by matching the performance of leading image-generation models using an oscillator-based architecture. Although currently simulated in software, Unconventional AI plans to release chip schematics and build a full inference stack. This approach could significantly alleviate the energy demands of AI, a critical challenge as the field continues to expand.
Read originalMore from TechCrunch AI
© TechCrunch AIMarket & Regulationmodels
White House Influences OpenAI's GPT 5.6 Release Strategy
OpenAI's latest model, GPT 5.6, is being released under unusual circumstances due to pressure from the Trump administration. Instead of a public launch, the model will initially be shared only with select partners, with the government approving access on a case-by-case basis. This cautious approach mirrors Anthropic's strategy with its Claude Mythos model, reflecting growing concerns over the potential misuse of powerful AI technologies. The administration's involvement marks a shift towards more federal oversight in AI development, highlighting the delicate balance between innovation and safety.
TechCrunch AI
© TechCrunch AIInvestment · $50M
Market & Regulationagents
Patronus AI secures $50M to enhance AI agent testing
Patronus AI is making waves with its innovative approach to testing AI agents in simulated digital environments. By creating 'digital world models,' the startup allows AI agents to be stress-tested in complex scenarios, ensuring they perform reliably in real-world tasks. This approach is akin to how autonomous vehicles are tested in synthetic worlds, highlighting its potential to revolutionize AI agent development. With a $50 million Series B funding round, Patronus is poised to expand its offerings beyond software engineering and finance, addressing more complex and non-verifiable problems.
TechCrunch AI
© TechCrunch AIMarket & Regulationbusiness
Anthropic's Claude Gains Traction Among Paid Consumers
Anthropic's AI model Claude is making significant inroads into the consumer market, traditionally dominated by ChatGPT. Data from Indagari shows a 75% increase in paying consumers for Claude since January 2026, highlighting its growing appeal. This surge is partly attributed to Anthropic's stance against using its models for mass surveillance, which resonated with consumers. While ChatGPT remains the leader, Claude's rapid growth in consumer interest and revenue signals a shift in the competitive landscape. As Anthropic and OpenAI approach potential public offerings, Claude's momentum could play a crucial role in shaping their market positions.
TechCrunch AI
More in Models & Labs
Models & Labsmodels
Llama.cpp b9784 Release Enhances Hexagon Performance
The latest b9784 release of llama.cpp brings significant optimizations to Hexagon's matrix multiplication capabilities. By reworking the MUL_MAT and MUL_MAT_ID operations, the update introduces a 32x32 tiled weight repack and improved kernel parameters, enhancing performance and efficiency. These changes aim to optimize register usage and streamline activation processing, particularly benefiting users leveraging Hexagon's architecture. This release doesn't introduce new models but focuses on refining existing processes, making llama.cpp more robust for developers working with diverse hardware configurations.
llama.cpp Releases
Models & Labsmodels
llama.cpp b9788 release enhances dual-GPU support
The latest release of llama.cpp, b9788, introduces significant improvements for dual-GPU setups with SYCL support, particularly enhancing tensor parallelism. By implementing a degenerate ring all-reduce for dual-GPU configurations, the update optimizes performance for both small and large tensor operations, mirroring CUDA's NCCL allreduce pattern. This release notably boosts performance metrics, with Llama-3.3-70B and Qwen3-Coder-Next-80B-A3B models showing substantial speed improvements. The update positions llama.cpp as a more competitive option for multi-GPU environments, without adding new dependencies or altering build configurations.
llama.cpp Releases
© The AI Daily BriefModels & Labsmodels
OpenAI Develops Custom Chip 'Jalapeño'
OpenAI has announced the development of its first custom chip, named 'Jalapeño'.
The AI Daily Brief