Model comparison

Performance and trade-offs

Vertical axis Horizontal axis Focus on observed range Show recorded 95% intervals

Compare a second metric

Grouped by model and instructions

Metrics shown

— = not measured. Intervals are shown where available. Cost and latency include their recorded measurement conditions below.

Inspect probability and threshold evidence

Exploratory development curves. These do not replace the comparison estimates or change a policy.

Candidate

Models, instructions and measurement conditions