Current proprietary flagship. 102B total / 12B active MoE with 128K context, optimized for Korean with English and Japanese support. IFBench: 55.78 (52% improvement over Pro 2), Arena Hard v2: 62.5 (30% improvement). Supports function calling, parallel tool use, and reasoning mode. Same throughput and cost as Pro 2. Base not explicitly disclosed, but the specs exactly match the from-scratch Solar Open 100B (same 102B/12B MoE, same SnapPO RL framework) — indicating the Pro line moved onto the in-house Solar Open lineage and off the earlier Phi/DUS lineage.

Model Details

Architecture MOE
Parameters 102B
Active params 12B
Context window 128,000
AA Intelligence 14
enterpriseagenticreasoningmoe

Related