Q4 REPORT
The Frontier Model Landscape: The Breakaway Q4 2025
This is our fourth edition of our quarterly report on the capabilities of generative AI models.
By 2025, generative AI models are embedded across the technology stack, and new releases increasingly shape industry discourse, investment, and enterprise procurement. Earlier quarters emphasized reasoning improvements and progress toward general intelligence. In Q4, the focus shifted from expanding general capability to hardening models for use inside real systems. Reinforcement learning and agent frameworks were applied to multi-step workflows where models must act, observe outcomes, and recover from errors. Coding, long a core use case, became the first domain where frontier labs optimized models for task completion across multiple dependent steps rather than single-response correctness. This report examines how that shift reshaped the frontier model landscape in Q4 2025.
