Large Language Model Performance Doubles Every 7 Months

Abstraction: LLM task-completion capability doubles every seven months exponentially

Key points:

• METR (Model Evaluation & Threat Research, Berkeley) published "Measuring AI Ability to Complete Long Tasks" (March 2025, arXiv:2503.14499)
• Key metric: "task-completion time horizon" — how long a human would take to do a task an LLM completes at 50% reliability
• Plot of this metric across major LLMs shows exponential growth with ~7-month doubling period
• Projection: by 2030, top LLMs should complete software tasks requiring a full human-month of work at 50% reliability
• "Messy" real-world tasks are harder for LLMs; messiness score correlates with lower performance
• Researcher Megan Kinniment notes hardware and robotics bottlenecks could slow otherwise explosive acceleration

Connections: Metr · Large Language Models · AI Benchmarking · Task Completion Horizon

Source: https://spectrum.ieee.org/large-language-model-performance?utm_source=techalert&utm_medium=email&utm_campaign=techalert-07-10-25&utm_content=httpsspectrumieeeorglargelanguagemodelperformance