ChatGPT Pro 500 vs a Sume wallet: what pays for a video job

A ChatGPT Pro 500 subscription raises how much you can use ChatGPT, Codex and the new Dots agents. It is a separate bill from the Sume wallet that pays for generated video, images, speech and music. If an agent running on a Pro 500 plan calls Sume, the clip is charged to your Sume workspace balance, not to the subscription, and the call still needs your Sume API key or MCP sign-in.

What OpenAI announced

The OpenAI Developer Community's DevDay 2026 roundup says Pro 500 subscriptions offer 25 times the usage allowance of Plus plus Ultrafast access, and that Pro 200 subscriptions were reopened, with new subscriptions differing from legacy grandfathered plans. It also lists Dots, persistent agents with connected apps, for Pro, Business and Enterprise tiers, excluding the European Economic Area, Switzerland and the UK at first.

The same roundup is about allowances inside ChatGPT. I did not find a statement that a plan includes API credit, so check OpenAI's own billing page for that. Token prices for the model are separate: the GPT-6.1 Sol thread lists $2 per million input tokens and $10 per million output tokens.

Three bills, three owners

When an agent makes media, up to three meters run. They do not offset each other.

What the Sume side needs

Sume's generation admission page says a paid submit fails with 402 insufficient_credits when Sume cannot reserve the estimated cost from the workspace balance. Concurrency is plan-based: the table lists Free at 1 processing slot up to Scale and Enterprise at 20, and prepaid top-ups do not raise it. A bigger ChatGPT plan does not raise either number.

If the agent calls Agent Completions, you must also send generation_spend_cap_usd; there is no default. The receipt's usage covers generation spend and excludes the agent's own LLM turn, so it is not the run's whole cost.

A practical budget rule

Budget the two sides separately. Decide how much a run may spend on media and set that as the cap; decide how much the model conversation may cost and watch it in OpenAI's dashboard. Always-on agents make this matter more, because no person is watching a run at 3 a.m.

• Give each unattended agent its own Sume API key from the dashboard so usage is attributable. See Authentication.
• Use dry_run or generation_admission_preview before expensive bursts.
• Check the balance before an overnight run, then cap each run with max_spend_usd or generation_spend_cap_usd.
• Do not paste your Sume key into a ChatGPT chat or a shared project file.