Question 1

How do I compare AI models on AIgateway?

Accepted Answer

Search the picker on /compare for the models you want, click each one to add a column, and AIgateway pulls live pricing, context window, and capabilities from each provider's docs. Shareable URL — copy the link and the comparison reproduces.

Question 2

How many models can I compare at once?

Accepted Answer

Up to 4 models side-by-side. Past that the columns get unreadable; for broader exploration use /models with filters.

Question 3

Where does the comparison data come from?

Accepted Answer

Each provider's published pricing and capability docs, plus our own catalog at https://api.aigateway.sh/v1/models which we keep in sync as new models ship. Cache reads, prompt caching, vision, and tool-calling flags reflect what the upstream provider actually supports.

Question 4

How does AIgateway pricing work?

Accepted Answer

Pass-through on the underlying provider rate plus a 5% platform fee at credit top-up — no markup per call, no monthly minimum. Cached requests get a 50% discount. New accounts get $5 signup credit (no card) on a curated edge-tier shortlist.

Question 5

Can I compare models across providers?

Accepted Answer

Yes — that's the point. Side-by-side any of 1000+ models from 85+ providers (Anthropic, OpenAI, Google, Moonshot, Meta, Mistral, xAI, Deepgram, ElevenLabs, Black Forest Labs, and more) without signing up for each.

Question 6

Can I run an eval to pick a winner on my own data?

Accepted Answer

Yes. POST /v1/evals with the candidate model slugs, a dataset, and a grader. AIgateway runs each prompt against each model and returns an alias that routes to the current winner. Re-run the eval when a new frontier model lands and your alias re-binds with zero code change.

Cosmos Predict 2.5 2B Distilled — and what?