GPT Image 2.5 Flare: 5 images a minute at Tier 1 vs Sume limits

OpenAI's Flare page lists 5 images per minute (IPM) at Tier 1, rising to 250 at Tier 5. Sume's docs describe different limits: plan concurrency, a queue, and 429 codes. They do not say that Sume mirrors OpenAI's tier table, so read the dashboard for your own numbers.

OpenAI's figures are from its model page; Sume's from the generation admission docs. Both read 2026-10-01.

What are OpenAI's Flare limits?

What limits does Sume apply?

Sume's admission docs list plan-based concurrency: Free 1, Pro 4, Startup 8, Scale 20, Enterprise 20. Concurrency is a dispatch limit, not a submit limit: extra jobs wait in a queue of max(3, concurrency x 5). The docs also say to prefer the generation_limits.concurrency_limit field in your dashboard over the static table.

What errors do I handle?

Three are documented: 429 queue_full when no accepted capacity is left (wait or cancel queued jobs, retry with the same idempotency key), 429 rate_limited for request volume (back off, use retry-after when present), and 402 insufficient_credits when the estimated cost cannot be reserved.

How do I send a burst?

Use mode: "async" so submits return a job at once, then read results from the job. Count the queue size from your plan before sending a batch; one that exceeds it gets queue_full rather than being slowed.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Idempotency-Key: flare-batch-0001" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "Studio photo of a ceramic mug on linen",
    "mode": "async"
  }'