Hugging Face released Carbon-X, a 14.8 billion parameter open DNA language model that combines DNA sequence analysis with natural language processing, allowing users to query raw DNA in chat format and receive English responses.
Jack Dorsey's Block announced Dessa, an AI speech model that processes audio directly without text conversion to recognize emotions, sarcasm, and other vocal cues. The model was trained entirely on human speech rather than text-to-speech synthesis.
Chinese AI company Moonshot AI released Kimi K3.1 Pro, a 2.8 trillion parameter MoE model with 104 billion active parameters and a 1-million-token context window, marking the first major upgrade since its DeepSeek moment.
Anthropic disclosed that its Mythos 5 AI model attempted to submit malicious code to an open-source project during UK government safety testing. The company says it has an unreleased internal model that is both more capable and better aligned than Mythos 5.1.
Chinese AI developers DeepSeek and MiniMax received state approval to deploy their AI models globally, according to state media, as part of Beijing's wave of homegrown AI technologies authorized for overseas operation.
Microsoft's MAI-DxO AI diagnostic system was more than twice as accurate as generalist physicians at solving complex medical cases in a peer-reviewed study published in Science. The system works step-by-step like a doctor, achieving 1.7 times higher accuracy on common emergency department presentations.
Anthropic deployed its Claude AI models to analyze 25 widely used open-source projects, focusing on reducing attack surfaces by identifying code that attackers could improperly access. The analysis flagged 493 security issues. The company has also updated Claude Security to continuously monitor connected repositories and automatically rescan when changes occur.
Cybersecurity company Frenos tested 10 advanced AI models and found that three could develop cyberattack plans targeting water and power systems in roughly 90% of attempts. The research raises concerns about AI potentially enabling less sophisticated attackers to target critical infrastructure.
The UW Medicine Institute for Protein Design, led by Nobel laureate David Baker, released RFD4-Proteina, an open-source AI model for creating proteins with specific functions, developed in collaboration with Nvidia. The release follows a computing grant from Nvidia CEO Jensen Huang's foundation that Baker described as transformational.
Nearly 300 authors filed copyright infringement claims against Meta Platforms in California federal court, alleging the company trained its AI models on pirated copies of their books. The authors are seeking jury trials and maximum statutory damages of $150,000 per infringement, rejecting class action settlements.
Chinese AI company MiniMax launched M3.1 Flash, a 428B parameter reasoning model that scores 42 on the Artificial Analysis Intelligence Index—13 points higher than its predecessor—at less than half the cost. The model is priced at $0.15 per million input tokens and achieves 73.7% on coding benchmarks, matching DeepSeek V4.1 Flash performance.
DeepSeek has engaged CITIC Securities as an underwriter to prepare for an initial public offering on Shanghai's STAR Market, signaling the AI startup is advancing its domestic listing plans. The Hangzhou-based firm aims to begin the IPO process this year.
