GPT-6.1 Sol Flex and Batch at half price: is there a Sume equivalent?

Not through the Format API. OpenAI lists GPT-6.1 Sol at half the Standard price on its Batch and Flex tiers and at twice the Standard price on Fast. A Sume Format run takes a model field and no service-tier or effort field, so you cannot request Batch or Flex pricing on a Format call. The Sume Agents picker has one tier control, a Fast checkbox, which the catalog code documents as a 2x service-tier option on GPT-6.1 Sol.

The nearest thing to batch work on Sume is a bulk run, which queues up to 100 Format runs. That is queueing, not a discount.

What are the vendor tiers?

The OpenAI pricing page lists the Standard, Batch, Flex and Fast rates per 1M tokens, each with a short-context and a long-context column. Short context here means a prompt up to 272K tokens.

What does Sume offer instead?

Two things exist on the Sume side. The Format call doc lists model as the field that selects the orchestrating LLM, and that is the only LLM control in the request body; a bulk-run item has the same body, so it has the same single field. In the Agents picker, the repository's catalog code gives GPT-6.1 Sol Low, Medium and High effort plus a Fast checkbox documented at twice Standard, and says the picker never offers a none level.

The picker is a chat control. It is not a way to set a tier on an API call, and the Format docs do not describe one.

Is a bulk run a Batch equivalent?

No, and the difference is concrete. OpenAI's Batch and Flex tiers trade latency for a lower price. A Sume bulk run takes 1 to 100 items and a concurrency window of 1 to 16, then runs the child Formats in order with that many in flight. It controls how many Format runs are active at once. It does not change what the LLM turn of each child costs.

If you need the half-price tier for a large text job, that is a direct OpenAI integration. If you need many Formats produced, use a bulk run and read the receipts.

How do I check what a run cost?

Both fields are described in the runs doc. Use the receipt, not a vendor list price, as the cost of a Sume run.

• Read usage.debited_usd_micros on each receipt: it is the wallet debit and includes the turn's own LLM row.
• Read usage.billable_amount_usd_micros for generation spend, which excludes the LLM turn.
• Subtract to see the LLM share per run.

What does half price look like in numbers?

Take 1M input tokens and 100K output tokens on the short-context card. Standard is $2 for input and $1 for output, $3 in total. Batch or Flex is $1 and $0.50, $1.50. Fast is $4 and $2, $6. The ratio between the three is the point: the cheap tier costs half, the fast tier twice.

None of that is a Sume price. It is what the vendor charges on its own API for the tier you pick there, and it is the reason a team with a large offline text job may prefer to call OpenAI directly for that job.

OpenAI's page also says the model has no audio or video input, only text and image. A video agent that needs a clip read passes frames or a description, which is a separate step from the model choice.