Prompt Studio

Run one prompt across models, effort levels, and thinking — side by side. Compare output, latency, and real token cost from the API's own usage. Pick a model on evidence, not vibes.

System (frozen · cached prefix)Shared across panes — the cacheable prefix.User promptUse {{variables}} to template inputs.

max_tokens

Fastest · cheapest · near-frontierthink

Output appears here.

Best speed ↔ intelligence balancethink

Output appears here.

Most capable · complex reasoningthink

Output appears here.