Baseline · Sol / Medium / Standard
Selected · Terra / High / Fast
Each lane keeps its native unit.
Terra / High / Fast: coding proxy −2.6 points; GeneBench −6.3 points at +7.8k average tokens; output rate +187.5 credits/M. No task-level winner inferred.
AA Coding Agent Index
Official model proxy · not effort-specific
No comparable OpenAI-published value0100 index
■ Baseline80
◆ Selected77.4
Δ −2.6 points
GeneBench-Pro score
129-problem biology eval · selected effort
No comparable model-by-effort value in this view0%35%
■ Baseline22.5%
◆ Selected16.2%
Δ −6.3 points
GeneBench average tokens
Trace + response tokens · same biology eval
No comparable model-by-effort token value in this view0k120k
■ Baseline14.4k
◆ Selected22.2k
Δ +7.8k tokens
ChatGPT output credits / M
Published rate card · Fast multiplier applied
Research preview · no token credit rate02,000 credits
■ Baseline750
◆ Selected937.5
Δ +187.5 credits
Model speed ordinal
OpenAI icon count · not a latency measure
No documented ordinal1 / 55 / 5
■ Baseline2 / 5
◆ Selected3 / 5
Δ +1 ordinal step
Fast service effect
Model speed only · not end-to-end time
Fast is not available for this model1.0× Standard1.5× model speed