Map/Measurement and understanding/Wave 9

AI R&D capability evals

Tests of how well models do the work of AI research itself.

Facts

Lane: Measurement and understanding
Wave: 9 of 12 (build order, not a date)
Needs: 2 boxes all the way back through hard links
Unlocks: 1 boxes downstream through hard links
Leverage: None
Cluster: Oversight and alignment

Evidence as of October 2026

METR’s time-horizon suite, including RE-Bench and HCAST, is the main public gauge of AI research ability. METR marked its public time-horizons page “no longer actively updated” on 8 September 2026.

01METR: time horizons

Around it

Needs · 3

Real-world task evalscapability · Emerging
Tool use and code executioncapability · Built today
Tests models cannot gamecapability · Emerging · helps, not required

Unlocks · 1

Internal deployment oversightcapability · Emerging

Loops and off-ramps

AI research automationreinforcing loop

Openings researchers' judgement, October 2026 scouting

Startup room 33/100 · rank 105 of 124 Research room 52/100 · rank 60 of 124

Crowding2 Lab overlap5 Small-team fit3 Buyer today2 Research openness4

Depth: light
Players: METR (RE-Bench, time horizons); UK AISI; frontier labs
Wedge: Measuring uplift on real lab R&D workflows before task suites saturate. research

01METR: AI R&D evaluation report