ServiceNow

An update to the Apriel series, focusing on using fewer tokens per answer while maintaining performance. They achieved this by using GSPO with length and verbosity penalties.

A transformer/mamba hybrid reasoning model from ServiceNow.

This model made the headlines mainly due to its high artificial analysis score, being on par with DeepSeeks models. However, this parity is mainly due to the differences in tau-bench. However, aside from that, it seems like a capable model.

ServiceNow is also joining the ever-growing list of companies training their own reasoning models.