Open AI models may be about a year behind closed AI models.

The claim, hedged with 'may be', is that the best open-weight AI models were about a year behind the best closed models as of January 2026. TRUE: credible benchmark-based estimates (for example Epoch AI, Artificial Analysis, LMArena) put the lag at roughly 9 to 15 months. MOSTLY TRUE: estimates put it at roughly 6 to 9 or 15 to 18 months. MIXED: credible estimates disagree widely, with some near a year and others far from it. MOSTLY FALSE: estimates put the gap under 6 months or between 18 and 30 months. FALSE: open models were at parity with or ahead of closed models, or more than 2.5 years behind.

Set before the research began.

1. E1 Open models lag state-of-the-art closed models by 4 months Epoch AI · 29 May 2026

   “Since January 2026, the most capable open-weight models have lagged frontier closed models by an average of four months in the Epoch Capabilities Index (ECI), our aggregate measure of model capability.”

2. E2 Open models lag state-of-the-art closed models by 4 months Epoch AI · 29 May 2026

   “This gap between open and closed models is slightly larger than the one we identified in our October 2025 Data Insight, which found that open models lagged by an average of three months between January 2023 and October 2025.”

3. E3 Open vs. closed AI: How behind are open models? Samaritan Research · 9 June 2023

   “In terms of benchmark performance, the best open large language models (LLMs) have lagged the best closed LLMs by 5 to 22 months.”

4. E4 Technical Performance Stanford HAI, AI Index 2026

   “As of March 2026, the top closed model leads the top open model by 3.3%, up from 0.5% in August 2024.”

Epoch AI estimates that since January 2026, the most capable open-weight models have lagged frontier closed models by an average of four months on its Epoch Capabilities Index [E1]. Its earlier analysis put the average lag at three months for January 2023 to October 2025 [E2]. Both figures fall in the criteria's 'under 6 months' band, not near a year. The Stanford AI Index gives only a percentage gap between the top closed and top open model, not a time lag [E4], and the 2023 Samaritan estimate of a 5 to 22 month lag predates January 2026 [E3].