Open-weights lag behind the closed frontier (months)
At each new open-weights record on Epoch's Capabilities Index, and at the newest model in Epoch's file, the months since a closed model first scored as high as the best open model to that date. Derived metric open_weights_lag_months; the time version of the open-vs-closed gap.
Reason through thisThe reading
Read at each open-weights record and at the newest model in Epoch's file, so the lag keeps growing through a stretch with no new record; the ECI points gap on the capture lens is the companion reading.
26 readings drawn from the derived metric open_weights_lag_months; 35 observations in the evidence.
Derived (open_weights_lag_months) from: Epoch AI benchmarks and Capabilities Index. Epoch AI, Benchmarking Hub and Epoch Capabilities Index (CC BY 4.0).
The numbers
| as of | reading | value | record |
|---|---|---|---|
| 2026-10-26 | — | 7.7 months | measured |
| 2026-10-17 | — | 7.4 months | measured |
| 2026-07-16 | — | 4.4 months | measured |
| 2026-06-16 | — | 6.9 months | measured |
| 2026-04-20 | — | 5.0 months | measured |
| 2026-04-07 | — | 8.0 months | measured |
| 2026-01-27 | — | 5.7 months | measured |
| 2025-12-01 | — | 7.5 months | measured |
| 2025-11-06 | — | 6.7 months | measured |
| 2025-09-29 | — | 5.5 months | measured |
| 2025-07-25 | — | 3.8 months | measured |
| 2025-05-28 | — | 5.3 months | measured |
| 2025-04-28 | — | 4.3 months | measured |
| 2025-01-20 | — | 1.1 months | measured |
| 2024-12-26 | — | 3.4 months | measured |
| 2024-12-12 | — | 3.0 months | measured |
| 2024-09-19 | — | 4.2 months | measured |
| 2024-07-23 | — | 2.3 months | measured |
| 2024-06-07 | — | 14.8 months | measured |
| 2024-05-07 | — | 13.8 months | measured |
| 2024-04-18 | — | 13.2 months | measured |
| 2024-04-17 | — | 13.1 months | measured |
| 2023-12-11 | — | 8.9 months | measured |
| 2023-11-02 | — | 7.7 months | measured |
| 2023-07-20 | — | 4.2 months | measured |
| 2023-07-18 | — | 4.1 months | measured |
Why it matters
How fast frontier capability becomes something anyone can download and run is the methods stock leaking into products outside the labs' control, and the capture lens's clearest sign of model-layer commoditisation.
What cuts against it
One composite index run by one organisation; open releases often land after internal use, so the lag at release overstates how long the capability was closed; a Chinese lab's release depends on compute export rules as much as on methods.
How it is measured
- derived metric
open_weights_lag_months(formula in the semantic layer) - source Epoch AI benchmarks and Capabilities Index · default tier 1 · CC BY 4.0
- proxy benchmark, model_release · cadence monthly · valve invention to product
Normal = open weights at least nine months behind, roughly a model generation; fast = three months or less, when a closed lead lasts about one release cycle. Between is emerging.
Applied to metric:open_weights_lag_months.
Derived rows (26)
| as of | dims | value | inputs |
|---|---|---|---|
| 2026-10-26 | 7.7 months | sources123 | |
| 2026-10-17 | 7.4 months | sources12 | |
| 2026-07-16 | 4.4 months | sources12 | |
| 2026-06-16 | 6.9 months | sources12 | |
| 2026-04-20 | 5.0 months | sources12 | |
| 2026-04-07 | 8.0 months | sources12 | |
| 2026-01-27 | 5.7 months | sources12 | |
| 2025-12-01 | 7.5 months | sources12 | |
| 2025-11-06 | 6.7 months | sources12 | |
| 2025-09-29 | 5.5 months | sources12 | |
| 2025-07-25 | 3.8 months | sources12 | |
| 2025-05-28 | 5.3 months | sources12 | |
| 2025-04-28 | 4.3 months | sources12 | |
| 2025-01-20 | 1.1 months | sources12 | |
| 2024-12-26 | 3.4 months | sources12 | |
| 2024-12-12 | 3.0 months | sources12 | |
| 2024-09-19 | 4.2 months | sources12 | |
| 2024-07-23 | 2.3 months | sources12 | |
| 2024-06-07 | 14.8 months | sources12 | |
| 2024-05-07 | 13.8 months | sources12 | |
| 2024-04-18 | 13.2 months | sources12 | |
| 2024-04-17 | 13.1 months | sources12 | |
| 2023-12-11 | 8.9 months | sources12 | |
| 2023-11-02 | 7.7 months | sources12 | |
| 2023-07-20 | 4.2 months | sources12 | |
| 2023-07-18 | 4.1 months | sources12 |
Latest readings
- 2026-10-26 · 7.7 monthsas of 2026-10-26(3 obs)
- 2026-10-17 · 7.4 monthsas of 2026-10-17(2 obs)
- 2026-07-16 · 4.4 monthsas of 2026-07-16(2 obs)
- 2026-06-16 · 6.9 monthsas of 2026-06-16(2 obs)
- 2026-04-20 · 5.0 monthsas of 2026-04-20(2 obs)
- 2026-04-07 · 8.0 monthsas of 2026-04-07(2 obs)
Why this status, and its history
The lag is now read at the newest model in Epoch's file as well as at each open record: 6.2 months at 9 Sep 2026, since a closed model first matched the best open score, set by Kimi K3 on 16 Jul 2026, on 5 Mar 2026. The earlier reason quoted 4.4 months, the lag on the day of that record, which stayed frozen while the true lag grew. It sits between the fast band (3 months or less) and the normal band (9 months or more), so the status stays emerging.
- 2026-09-23emerging to emergingconf 45 → 45 · Claude (model)
The lag is now read at the newest model in Epoch's file as well as at each open record: 6.2 months at 9 Sep 2026, since a closed model first matched the best open score, set by Kimi K3 on 16 Jul 2026, on 5 Mar 2026. The earlier reason quoted 4.4 months, the lag on the day of that record, which stayed frozen while the true lag grew. It sits between the fast band (3 months or less) and the normal band (9 months or more), so the status stays emerging.
Weighed against it: One composite index run by one organisation; open releases often land after internal use, so the lag at release overstates how long a capability was closed.
sources123 - 2026-09-11unmeasured to emergingconf — → 45 · Claude (model)
First reading: at its latest record (16 Jul 2026 on Epoch's Capabilities Index) the open-weights frontier trailed the closed frontier by 4.4 months, between the fast band (3 months or less) and the normal band (9 months or more), so the status is emerging until the lag settles on one side.
sources12
Why this confidence
Rests on 35 observations from Epoch AI benchmarks and Capabilities Index; best evidence tier 1, grade A; current.
Confidence is independent of status: 90–95 multiple strong independent sources; 70–89 good evidence, some ambiguity; 50–69 mixed or hard to operationalise; 0–49 limited or vague evidence.
Predictions that lean on this
- Happening
- Both sides expect this