We are excited to announce Day-1 availability of Open AI’s first open model, GPT-OSS-120B, as well as a new vision model, Qwen-2.5-VL. You can try these models out by logging onto SCRP-Chat.

GPT-OSS-120B

This new model, released by Open AI on Aug. 6th 2025, is roughly on par with the their proprietary o4-mini model. In our internal test, the model is very strong in mathematics and reasoning, while being significantly faster than the other models we offer.

Max. Throughput: 155 token/s

For more details on the model, please refer to OpenAI’s announcement and model card:

Qwen-2.5-VL

This model is significantly better in Chinese recognition than our previous vision models, Pixtral-Large and Gemma 3 27B. While the model has text capability, it is prone to getting stuck in repetition for tasks that involve long reasoning. As such, we do not recommend using this model if your task involves complex mathematics or reasoning.