Skip to content
AI4UAtlasChat

Solar Open2 250B

28 June 2026

Solar Open2 250B from Upstage, language model published 2026-06-28. Notable for: training cost,discretionary.

Connections

Developed by 1

Abstract

Solar Open2 250B is an open-source Mixture-of-Experts (MoE) model developed by Upstage. It features 250.3B total parameters with 14.7B active parameters per token. The model is trained on approximately 11 trillion tokens using NVIDIA B200, H200 GPUs. Supporting a context length of up to 1M tokens, it is designed to excel in agentic tool use, long-context retrieval, instruction following, and reliable, low-hallucination reasoning, with strong multilingual performance particularly in Korean, English, and Japanese.

Sources: Epoch AI. This entry was generated from a dataset and has not been edited by hand.