February 25, 2025 News

Anthropic's Claude 3.7 Sonnet Cost Only Tens of Millions to Train

According to information reportedly provided by Anthropic to Wharton professor Ethan Mollick, their latest flagship AI model Claude 3.7 Sonnet cost only "a few tens of millions of dollars" to train using less than 10^26 FLOPs. This relatively modest training cost for a state-of-the-art model demonstrates the declining expenses of developing cutting-edge AI systems compared to earlier generations that cost $100-200 million.

OpenAI Delays API Release of Deep Research Model Due to Persuasion Concerns

OpenAI has decided not to release its deep research model to its developer API while it reconsiders its approach to assessing AI persuasion risks. The model, an optimized version of OpenAI's o3 reasoning model, demonstrated superior persuasive capabilities compared to the company's other available models in internal testing, raising concerns about potential misuse despite its high computing costs.

DeepSeek Resumes API Services After Capacity-Driven Pause

Chinese AI startup DeepSeek has reopened access to its API after a three-week pause caused by capacity constraints. The company's openly available R1 reasoning model has gained recognition for matching or exceeding the performance of OpenAI's top models, prompting competitive responses from both OpenAI and domestic rivals like Alibaba.

Microsoft Reduces Data Center Footprint Amid Industry Expansion

Microsoft has canceled leases totaling approximately two data centers' worth of capacity with multiple providers. This contraction stands in contrast to significant industry expansion, including the $500 billion Stargate project by OpenAI, Oracle, and SoftBank, and raises questions about Microsoft's expectations for future AI demand.

Koyeb Integrates Tenstorrent's RISC-V AI Accelerators into Serverless Platform

Cloud platform Koyeb has deployed Tenstorrent's AI accelerators, offering developers access to an alternative to Nvidia's GPUs. This partnership follows Tenstorrent's recent $700 million funding round and represents part of a broader effort to build hardware and software alternatives to Nvidia's dominant AI stack.