Model · published numbers Fast · published multiplier Effort · one domain matrix
Coding signal
77.4
−2.6 points vs Sol
Published speed
3 / 5
Standard service tier
API list price / 1M
$2.50 / $15
input / output · 50% below Sol
High suits difficult work with multiple steps, sources, or tradeoffs.
GeneBench-Pro example: Terra high scored 16.2% with 22.2k average tokens — +2.6 points and 1.40× tokens versus Terra medium.
Ultra is a parallel multi-agent mode, not another reasoning-effort notch.

Release benchmark profile

Sol marker │ selected dot ●
AA Coding Agent Index
index points
77.4 −2.6
Terminal-Bench 2.1
pass rate
87.4% −1.4
Agents’ Last Exam
score
50.4% −2.3
OpenAI release results are model-level reference points, not a prediction for the selected effort or Fast mode.
Benchmarks Model guidance Speed API pricing Effort matrix