Solar Decide

Last updated:

WrapperProprietary

⚠️ Why not a model

Upstage serves it on Solar Mini 4 and reads the decision from the logits of single-token labels A–Z. No decision-specific weights, head or model card documented, and no independent benchmark.

Technical specs

Base LLMs
Solar Mini 4
Decision types
choice, score, noul
Features
hosted-api, beta, systemone-endpoint, openrouter, single-token-labels, max-26-options, zero-data-retention
License
Proprietary

Solar Decide is a beta decision endpoint from Upstage. Its console page documents POST https://api.upstage.ai/v1/systemone, says the service is “served on Solar Mini 4” and that it “reads the decision from the logits of single-token labels A through Z”. The same product is listed on OpenRouter as upstage/solar-decide, created on 28 September 2026.

What is public

As of 2 Oct 2026
Endpoint POST https://api.upstage.ai/v1/systemone (beta); OpenRouter upstage/solar-decide
Engine Solar Mini 4 (Upstage’s general LLM)
How it decides Probabilities from the logits of single-token labels A–Z, so at most 26 options per question
Question types Choice, Noul, Score in the System One request shape
Context 524,288 tokens (OpenRouter listing)
Price $0.05 / M input on OpenRouter, shown as a 50% launch discount
Data Zero data retention flag on OpenRouter

Why it is listed here (not in Models)

The catalog admits models with trained decision weights (criteria). Upstage documents a readout method on top of a general LLM, not a trained decision model: there is no model card, no decision head and nothing saying Solar Mini 4 was tuned for this. Reading label-token logits is the same technique open wrappers use on any LLM. It is listed as wrapper and stays off the quadrant.

Re-evaluation trigger: a model card or announcement that describes decision-specific training, or an independent benchmark (JevBench, Decision Index) on the endpoint.

Vendor numbers (UnverifiedClaim)

None with a source. Upstage has published no latency, accuracy or calibration figures for Solar Decide that we could find.