Inception Mercury Decide: A Structured Decision Model
Inception released its structured decision model Mercury Decide on 30 September 2026: the model card says it is not a chat model but a System One endpoint where you send a state along with typed questions and get back a choice, a score or a yes/no answer with a calibrated probability taken directly from the model, which is why output tokens are free; it makes up to 14 decisions per second and reports how certain it is, so a decision system can run it on every case and escalate only the unsure ones to a human, using the same /v1/systemone schema as Jev
Tool Overview
Features, steps and FAQ below
Features
- ✓ Typed, structured output: you send a state (string, object or array) plus questions, and it returns a choice between options, an ordered score, or yes/no probabilities as machine-readable values instead of prose
- ✓ Calibrated probabilities: the model card says the probability comes straight from the model rather than being written out as text, so output tokens are free and callers can set thresholds to decide what runs automatically and what gets escalated
- ✓ High throughput, low latency: it is documented at up to 14 decisions per second, and OpenRouter's model page lists a p50 latency of about 0.32 seconds, which suits routing, ranking and validation decisions that sit in the hot path
- ✓ Interface shared with Jev: the model card says Mercury Decide uses the same /v1/systemone schema as TypeSafe's Jev, so the same call and parsing logic can be reused
- ✓ Free early access plus a cheap endpoint: OpenRouter hosts a free mercury-decide:free variant, with a discounted endpoint at $0.02 per million input tokens and free output, on a 66K context window
How to Use
- Pick Mercury Decide on Inception's provider page on OpenRouter (a free variant is available during early access)
- Note that it runs on the OpenRouter Decisions API, not the OpenAI-compatible chat completions endpoint
- Send the current state together with your typed questions (choice, score or yes/no)
- Read back the choice, score and calibrated probability, then route, rank or escalate the unsure cases to a human
FAQ
What is Mercury Decide?
Inception's structured decision model, which its model card describes as served as a System One endpoint that trades typed questions for a choice, a score or a yes/no answer with a calibrated probability attached.
When was it released?
OpenRouter's model page lists the release date as 30 September 2026.
Why is output free?
The model card says the probability comes directly from the model instead of being written out as text, so no output tokens are billed.
Can it hold a conversation like a chat model?
No — it does not generate text, and is positioned for structured decisions such as routing, ranking and validation rather than chat.
How do I call it?
It uses the same /v1/systemone schema as Jev through the OpenRouter Decisions API, so standard chat completions SDKs will not work with it.