Upstage
Upstage has published a first-party blog post announcing Solar Pro 3 as the March 2026 update to its Solar Pro large language model line. The post states that Solar Pro 3 delivers roughly 2× gains over Solar Pro 2 across every evaluation category, with the headline improvement being agentic performance — Tau2-all rises According to the same announcement, the agentic improvements are driven by SnapPO, Upstage's proprietary in-house reinforcement learning framework developed through the Solar Open model research program, which is designed to compose independent training stages so that reasoning can be improved simultaneously across mat