• Input / Output are the pay-as-you-go prices per one million tokens.
• The est./answer columns estimate one assistant turn on each plan — roughly 68K tokens of inlined docs context plus a ~600-token reply (capped at each model’s context window). Input dominates, so a single answer costs far less than the per-million output rate suggests; your real numbers vary with your docs size and answer length. On a paid plan, answers drain the plan’s included monthly allowance before anything is billed.
• Quality is the independent Artificial Analysis Intelligence Index (higher is better); it’s blank for models Artificial Analysis doesn’t cover.

A model priced $0.00 runs on a shared upstream free pool — handy for trying things out, but with real trade-offs you should plan around:

• Slow and queue-bound. Free requests are low priority and share a queue, so responses are slower and degrade further under load.
• Capped per day and per minute. Free usage is limited to roughly a few dozen requests per day plus a low per-minute rate; once you hit the cap, requests are refused until the window resets.
• Best-effort availability. Free endpoints can be throttled, deprecated, or briefly unavailable with no notice.
• Your inputs may train models. Some free providers may use prompts for training, so don’t send anything sensitive through a free model.

For a public Ask AI assistant under real traffic, choose a paid model — the limits above will otherwise rate-limit or stall your readers.