Latest AI News

Google DeepMind's and OpenAI's AI models achieved gold medal-level scores at the 2025 International Mathematical Olympiad, a first for artificial intelligence. Both systems solved five of six problems, scoring 35 out of 42 points while following the same two-day competition format as human participants. Unlike previous approaches, these models processed mathematical problems directly in natural language, producing human-readable proofs without requiring translation to machine formats. This breakthrough suggests AI could soon become a regular collaborator in mathematical research, demonstrating significant advancement in complex reasoning capabilities.

OpenAI and SoftBank's Stargate project, initially a $500 billion plan for massive AI data centers across the US, has scaled back its immediate ambitions due to partnership tensions. The first facility in Abilene, Texas spans 900 acres with Oracle supplying infrastructure and Crusoe securing $11.6 billion in funding. Instead of immediately building multiple centers, they now aim to have one operational by late 2025. Meanwhile, OpenAI has formed separate agreements with Oracle and CoreWeave for computing resources, while global expansion continues with plans for a UAE facility in 2026. Despite setbacks, the initiative remains significant for the future of AI infrastructure.

AI models from Google and OpenAI achieved gold medal scores at the 2025 International Math Olympiad. Both models scored 35 out of 42 points by solving five of six complex math problems. This marks a significant breakthrough in AI reasoning capabilities as these models processed natural language problems without translation. Google's Gemini Deep Think was officially recognized while OpenAI's model was evaluated independently.

AI systems from Google and OpenAI have achieved gold medal-level scores at the 2025 International Mathematical Olympiad (IMO), a prestigious competition for high school students. Google's Gemini Deep Think model officially scored 35 out of 42 points, solving five of six problems perfectly. OpenAI's experimental model achieved the same score, though its results were independently verified rather than officially certified by IMO judges. Both systems followed human contestant rules during the competition. This achievement marks a significant breakthrough in AI's mathematical reasoning capabilities, though debates continue about benchmarking consistency between the two systems. Only 10% of the 630 human contestants earned gold medals, highlighting the elite level of the AI's performance.