What would make this model API page more useful before integration?

Good question! And the instinct to make these decision-oriented rather than just call-docs is spot on. I'm not the deepest engineer, I build with these models day to day across a fleet of bots, so my answer is from the "will this actually do the job" side rather than the benchmarks side.
The thing I most want before integrating isn't the API mechanics — it's an honest "what's this bad at." Everyone lists what a model is good at. Almost nobody tells you where it falls over, and that's the bit I only find out once it's live and already wired in. A page that said "great for X, don't bother for Y" would save me more time than any amount of spec detail.

Second would be a realistic example for the kind of job I'm actually doing. A generic chat example tells me nothing about whether it'll hold up in, say, a support-triage loop or an agent that has to take actions. The workflow type matters more than the raw capability.
On your A/B/C — I'd start with B (use cases, and crucially when NOT to use it). Pricing matters, but only after I've worked out whether it can do the job at all. "Is this the right tool" first, "what does it cost" second.

Finally, looking at the actual page — the Model Type filter (Text / Image / Video / Audio) is probably too blunt to be the main way in. For the kind of work I do, "text generation" covers everything from a model that can barely follow an instruction to one that can run a reasoning-heavy triage loop unsupervised — and the gap between those is the entire decision. The modality tells me almost nothing; what I actually need to filter on is "can it think well enough for this job." I'd care far less about Text vs Image than about reasoning, instruction-following, and how it behaves in an agent/tool-use loop.

Hope the feedback helps! Good luck!