Main finding
Epoch AI reports that, since January 2026, the best open-weight models have trailed frontier closed models by an average of four months on the Epoch Capabilities Index (ECI), the organization's aggregate measure of model capability. As Epoch AI puts it: 'The average ECI gap was 8 points, similar to the gap between GPT-5 and GPT-5.5.'
Comparison to prior estimate
That gap is slightly wider than the three-month lag Epoch found in an earlier analysis published in October 2025, which covered the period from January 2023 to October 2025.
Methodology
To compute the time gap, Epoch AI examined its analysis window day by day from January 1 to May 28, 2026. For each day, it identified the top-scoring open-weight model available by that date and asked: what is the most recent date on which no closed-weight model was significantly better than this open model? The number of days since that date is the time gap for that day. Because ECI scores carry statistical uncertainty, Epoch used bootstrap resampling of its benchmark data to make the comparisons, treating an open model as having caught up if it beat a prior closed state-of-the-art model in at least 5% of paired bootstrap samples. On that basis, the average time gap came out to four months; requiring the open model's point estimate to be strictly higher (rather than just non-significantly worse) would push the estimate to six months. Measuring the gap in ECI points rather than time — the average vertical distance between the closed and open frontiers across the window — came out to 8 points, with a 90% confidence interval of 7 to 11 points.
Limitations
Epoch AI flags two reasons its estimate may understate the true gap: open-weight models tend to score worse, relative to closed models, on private benchmarks than on public ones — plausibly because developers optimize more aggressively against public benchmark leaderboards; and the analysis only includes models with enough public benchmark coverage to compute an ECI score, while leading closed labs do not always release their most capable models publicly, for safety, commercial, or competitive reasons.
Sources
- Open models lag state-of-the-art closed models by 4 months — Epoch AI