Returns typed choices, scores and yes/no answers with model-derived probabilities in one forward pass, instead of generated text.
Output Speed *
Intelligence Index *
Context Window *
Input price
Output price
Performance
Single-Pass Decisions with Probabilities You Can Threshold
Technical Specifications
What the model supports
Comparison
Solar Decide or Solar Decide Flash for Routing and Policy Checks?
Solar Decide and Solar Decide Flash are the closest comparison because both are System One decision models built on Solar Mini 4. Both return typed answers with probabilities in a single forward pass, keep the 512K context window, and list the same input rate with free output tokens on OpenRouter. Upstage describes Flash as the faster, low-latency variant tuned for consistent response times. The decision is whether response-time consistency matters more than staying on the baseline model.
Choose Solar Decide when you want the baseline decision model, are validating a decision pipeline in beta, or run asynchronous jobs such as batch document review where tail latency matters less. Choose Solar Decide Flash when decisions sit in a synchronous path, such as live ticket triage or request gating, and consistent response times matter. No published accuracy comparison between the two exists, so run both on your labeled data before committing.
Limitations & Trade-offs
Best-Fit Workloads
Where this model earns its place
Support ticket routing
Upstage's own example sends a customer ticket as the state and asks typed questions about it. One call can decide the queue (up to 26 options), whether the customer wants a refund (yes/no), and an urgency score. Each answer arrives with a probability, so low-confidence tickets can go to a person. The 524K context window also covers long ticket threads. It does not write the reply.
Long-document policy and compliance checks
Upstage positions the model for policy checks, and the 524,288-token context lets an entire document serve as the state. A contract or policy file can be tested against yes/no questions, such as whether a clause breaks a rule. The returned probability supports review thresholds. Because it produces no explanation, flagged items still need a reviewer or a generative model to cite the relevant passage.
Korean-language classification and moderation
Upstage says Solar Decide brings Solar Mini 4's strong Korean understanding to routing, classification and policy checks. That makes it a candidate for Korean support content, community moderation and intent detection, where yes/no or category outputs are enough. No independent Korean benchmark for this endpoint has been published, so validate it on your own labeled Korean data.
Confidence-gated agent routing
In an agent pipeline, Solar Decide can act as a gate before expensive model calls. It can pick a route, check a guardrail condition or score a request, then escalate only when the returned probability falls below a threshold. A single forward pass per question suits this role. It is a decision component, not a planner, and it does not call tools.