Free & sponsored models

• A model with zero token prices but a per-request fee or a minimum charge is paid, and is never labelled Free.
• Marketplace (seller) endpoints are never covered by a sponsorship.
• Free use is always limited, because the provider's free quota is shared by every Ditfa customer. Limits can apply per user, per key, across all users and to requests in flight. Model pages show an offer's dates, eligibility and per-user limits.
• An offer can cap the output of each request. A smaller max_tokens you send stays in force.
• A free or sponsored request places no hold on your wallet and does not count against a key's spend limit. Allowed models, IP allowlists, expiry and provider policy still apply.
• Every successful response carries x-ditfa-charge-mode with paid, free or sponsored.

{
  "error": {
    "message": "Free access limit for this model is used up. It resets at 2026-11-01T00:00:00.000Z. To use it at the normal price from your wallet, retry with the header \"X-Ditfa-Allow-Paid: true\".",
    "type": "rate_limit_error",
    "code": "free_quota_exceeded",
    "param": null
  }
}

When a limit has a reset time, the message says when and the x-ditfa-free-reset header carries it as an ISO 8601 time. A failed upstream call does not use up your limit.

curl https://gateway.ditfa.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DITFA_API_KEY" \
  -H "X-Ditfa-Allow-Paid: true" \
  -d '{"model": "openai/gpt-4o-mini", "messages": [{"role": "user", "content": "Hello"}]}'

• The opt-in only counts after a refusal. If the offer can serve the request, it stays at $0.
• The ditfa field is removed before the request reaches the provider.
• A paid continuation is an ordinary paid request: the wallet hold, insufficient_credits and key_limit_reached apply.
• It needs a paid endpoint to continue on. A model with only free endpoints keeps refusing, and so do wallet_frozen and account_disabled.
• The playground asks you each time.