New AI model beats previous benchmarks in reasoning tasks

A new language model was announced this week with results that surpass leading competitors on logical reasoning, math, and coding benchmarks.

According to its developers, the gains come from a new training technique that combines reinforcement learning with high-quality synthetic data, allowing the model to 'think' longer before answering complex questions.

Companies across several industries have already started testing the new technology in internal workflows, including customer support, code generation, and legal document analysis.

Experts point out that the race for increasingly capable models is expected to keep accelerating in the coming months, with multiple labs promising new releases later this year.