Markov and the companies nearest it
Grouped on reinforcement learning, data labeling, data engineering — sectors these companies actually share, not a category everything is in.
The full picture
- It follows you to every device, instead of living in this browser.
- Everything else you save from these pages lands in the same list.
- Nothing is emailed to you unless you switch it on yourself.
Saved here on this device in the meantime, so nothing is lost either way.
Markov
Expert computer-use data for frontier AI labs
Markov
New accelerator entry · 2026-06-24 · the record
Against the 209 companies on file from Summer 2026. Median team 2, largest 30.
CoArena
The biggest crowdsourced benchmark for Computer-Use
CoArena
New accelerator entry · 2026-06-24 · the record
Against the 209 companies on file from Summer 2026. Median team 2, largest 30.
CueBench
RL post-training for science reasoning and performance engineering
CueBench
New accelerator entry · 2026-08-14 · the record
Against the 209 companies on file from Summer 2026. Median team 2, largest 30.
Olam Labs
Building multi-agent simulations for model evals and training.
Olam Labs
New accelerator entry · 2026-07-20 · the record
Against the 209 companies on file from Summer 2026. Median team 2, largest 30.
Magma
Monetize your agent's traces.
Magma
New accelerator entry · 2026-08-13 · the record
Against the 209 companies on file from Summer 2026. Median team 2, largest 30.
Moving Atoms
World Models for training and evaluating robots
Moving Atoms
New accelerator entry · 2026-08-11 · the record
Against the 209 companies on file from Summer 2026. Median team 2, largest 30.
Alkera AI
Reliable and safe data engineering and data science agents
Alkera AI
New accelerator entry · 2026-07-10 · the record
Against the 209 companies on file from Summer 2026. Median team 2, largest 30.
Litmus
Evals for humans
Litmus
New accelerator entry · 2026-07-08 · the record
Against the 209 companies on file from Summer 2026. Median team 2, largest 30.
Use the arrow keys, or swipe, to move between companies. Every figure is read from the source record each company links to.
A deck arranges pages that already exist; it states nothing the individual profiles do not. Figures are read directly from the source record; names have not been matched against other people.