Format runs with no model field now run GPT-6.1 Sol, not gpt-6-sol

If you send a Format run without a model field, the orchestrator is now GPT-6.1 Sol (gpt-6.1-sol), not gpt-6-sol. That is what Sume's code on current main does, and it is the reverse of what the Create a run table still says, so do not trust either: read model on the run receipt.

The change landed on 2026-09-30 with the GPT-6.1 Sol row, and a later change made the default apply on every environment and harness. Nothing about the request shape changed. A Format run still takes instruction, input, output_schema and a spend cap exactly as before.

Which model does a Format run use when model is omitted?

Two places in Sume disagree today, and it helps to name both.

The docs page for Create a run says to omit model for the gpt-6-sol default and that a request for retired gpt-5.6-sol runs on gpt-6-sol. The server code that resolves a create call says the opposite for the omitted case: when the caller named no model, the run is stamped with gpt-6.1-sol, as if the caller had named it. The code comment also says the run takes the same Codex harness the older Sol pin used, so only the model id changes, not the runner.

Retired ids are a separate path. A caller who sends gpt-5.6-sol explicitly is still moved onto gpt-6-sol, as the retirement post describes. Only a request that names nothing gets 6.1.

How do I confirm which model ran my Format?

The receipt carries a model field, documented in Runs and results as the catalog id the orchestrator ran on. Fetch the result URL from the create response and print it.

Log that value next to your own order or request id. If the docs table and the receipt ever disagree again, the receipt is what ran and what you were billed on.

curl -sS "https://api.sume.com/v1/format-runs/$RUN_ID/result" \
  -H "Authorization: Bearer $SUME_API_KEY" | jq -r '.data.model'

Is GPT-6.1 Sol different from GPT-6 Sol for a Format?

On OpenAI's own pages the two are close. Both list a 1,050,000-token context window, 128,000 max output tokens and $2 per million input tokens and $10 per million output tokens. OpenAI describes GPT-6.1 Sol as near-Astra performance for complex work at a lower cost, and GPT-6 Sol as built for complex coding and agentic workflows.

The differences that show on those pages are small and mostly about the knobs:

• OpenAI list prices are context only. What a Format run costs you is on the receipt's usage object, under Sume's own rates and your generation_spend_cap_usd.
• The orchestrator picks tools and writes prompts. The image, video and audio models still come from the Format's own tools, so this change does not move which video model renders.

How do I keep gpt-6-sol or choose another orchestrator?

Send the id. {"model": "gpt-6-sol"} keeps the older Sol, because Sume lists GPT-6 Sol as its own row and stored GPT-6 Sol picks stay on it. gpt-6.1-sol is accepted too, and an id outside the catalog is 400 invalid_request, per the errors page.

Pin when you compare cost or latency between two runs, because an omitted field can move again when a new Sol ships. Leave it off when you want Sume's current default and you log the receipt.

What does this not change?

It does not change idempotency, webhooks or the output schema contract. It does not make Sume disclose a model for media, which is a separate question covered in which model did my AI video use. And it does not let you pick the LLM on every surface: Agent Completions accepts only model: "sume-agent".