OpenAI o4-mini
OpenAI
Too recent for its influence to be judged: no later milestone in the data descends from it yet. How levels are decided
Figures quoted here are reported by the organisation that built it, not independently verified.
Full summary in progress. Until then, this page shows the milestone's record from the research data.
Reasoning models were expensive enough to reserve for hard problems.
Agents make many calls, so per-call reasoning cost compounds badly.
A small reasoning model cheap enough to call repeatedly, with tool use.
Cheap reasoning is what makes long agent loops economically possible.
Reasoning at a price point agents can afford.
The same logic drives the mini and flash tiers across all labs.
The milestones it descends from, and the milestones that descend from it. Work developed alongside it is not counted. Weak links are left out unless you include them.
No documented ancestors are recorded for o4-mini without weak links.
No documented ancestors are recorded for o4-mini without weak links.
No recorded ancestors are recorded for o4-mini.
No recorded ancestors are recorded for o4-mini.
No documented descendants are recorded for o4-mini without weak links.
No documented descendants are recorded for o4-mini without weak links.
No recorded descendants are recorded for o4-mini.
No recorded descendants are recorded for o4-mini.
Developed alongside
None recorded in the sources used.