Solar Pro 3 0323
- Source
- upstage.ai
- Author
- Upstage
- Date
Context
Upstage's March 2026 Solar Pro 3 update targets a problem the company says it kept seeing in production: individual tool calls succeed while the end-to-end task never finishes. Upstage attributes the improved multistep reasoning to SnapPO, its proprietary reinforcement-learning technique, and reports a score of 72.3 on Tau2-all, an , against 36.0 for Solar Pro 2.
Upstage also says the update keeps the same API interface and throughput, and that existing Solar Pro 2 users can switch with no configuration changes. If accurate, that makes it easy to test in existing agent workflows. The roughly doubled score is Upstage's own reported result; it does not quantify improvement in any customer's workflow. Comparison and operating claims.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
- mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
Checking sign-in…
Loading comments…