Solar Open2 250B

Training compute

1.1×10²⁴ FLOP

Parameters

250.3B

Published

Jun 28, 2026

Solar Open2 250B is an AI model developed by Upstage (South Korea), first published in June 2026. It works in the language domain, on tasks such as language modeling, reasoning, tool use and 2 more.

Training it took an estimated 1.1×10²⁴ FLOP of compute. The model has 250,300,000,000 parameters. It was trained on roughly 11T datapoints. Training ran on 1,656 NVIDIA B200,NVIDIA H200 for about 1.9K hours.

Access: Open weights (unrestricted). Its weights are openly available. Epoch AI rates the confidence of this record as likely.

Full record

Organization: Upstage
Country of organization: South Korea
Domain: Language
Task: Language modeling, Reasoning, Tool use, Translation, Question answering
Training compute: 1.1×10²⁴ FLOP
Parameters: 250,300,000,000
Dataset size: 11T
Training hardware: NVIDIA B200, NVIDIA H200
Chips used: 1,656
Training time: 1,900 h
Training power draw: 3.2 MW
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Likely