THE FRONTIER SIGNALSIs AI progress accelerating exponentially?
Selected capability evaluations show improvement, but a curated timeline cannot establish an exponential industry-wide growth rate.
Open interactive Timeline hub →
Which milestones are documented?
- Colossus infrastructure documented 2024-10-28
- AlphaGenome introduced 2025-06-25
- AMD–OpenAI agreement 2025-10-06
- METR revises its time-horizon suite 2026-01-29
- DeepSeek V4 preview 2026-04-24
- Microsoft–OpenAI terms amended 2026-04-27
- DeepSeek V4-Pro GA 2026-08-13
- AlphaGenome Atlas announced 2026-09-08
- Claude Opus 5.5 introduced 2026-09-22
- METR publishes Opus 5.5 assessment 2026-09-22
- AI-assisted enzyme finding reported 2026-09-23
Frequency in this curated collection is not a test of exponential growth.
What else do readers ask?
Are AI releases becoming exponentially more frequent?
This curated collection cannot establish that. It lacks stable exhaustive coverage and a fixed event definition across time.
Is release frequency the same as capability growth?
No. More launches can reflect packaging, market competition or reporting changes without equivalent capability gains.
What does the METR chart actually measure?
It estimates human task duration associated with a chosen AI success rate on a selected software and research task suite.
Why are confidence intervals so wide?
Estimates depend on a limited, varied task sample and a fitted relationship. Uncertainty must remain visible, especially near measurement limits.
Why not fit an exponential curve to the milestone counts?
A biased, incomplete sample can create a misleading trend; a credible fit requires consistent coverage and comparable windows.
What evidence would make acceleration more convincing?
Repeated gains on stable, independently evaluated tasks, with comparable budgets and uncertainty, would be stronger evidence than headline density alone.