Integrate

Your capacity is live.Change the baseURL and carry on.

An OpenAI-compatible client. No new library, no code change beyond the base address.

Base address

https://qa.fullcup.ai

Same route, same fields, same responses as the client you already use.

Your key

fc_live_••••••••••••••••••••••••

Shown once, in the email. Lost it? Reissue the key; the previous one stops working at once.

Reissue the key in the studio

First call

curl https://qa.fullcup.ai/v1/chat/completions \
  -H "Authorization: Bearer $FULLCUP_KEY" \
  -H "Content-Type: application/json" \
  -H "X-FullCup-Workload: document-translation" \
  -d '{"model":"@cf/openai/gpt-oss-120b",
       "messages":[{"role":"user","content":"hello"}],
       "max_tokens":2048}'

X-FullCup-Workload: a free label you choose for each kind of work, such as document-translation. It lets a backer see what the capacity was spent on.

Token ceiling and reasoning models

Four of the five top models return empty content when max_tokens is low: reasoning eats the ceiling before the answer. If your first call comes back empty, raise max_tokens. It did not fail.

Models allowed in your class

Essential since Oct 9, 2026

@cf/meta/llama-4-scout-17b-16e-instruct
@cf/google/gemma-4-26b-a4b-it
@cf/openai/gpt-oss-120b
@cf/qwen/qwen3-30b-a3b-fp8
@cf/zai-org/glm-5.3-flash

A model outside the class is refused before any capacity is spent. The list is dated: what counts is the version in force when the cup was funded.

Consumption is public by design.

Every request emits a signed receipt. Your public cup page shows tokens, tag and time, with no login, in about a minute. This is not fine print: it is why anyone trusts you with capacity.

See a consumption page

POST /v1/chat/completions · GET /v1/models