Together AI Launches DeepSeek-V3.1: A Versatile Hybrid Model

Together AI has introduced DeepSeek-V3.1, a hybrid model designed to balance speed and depth in reasoning tasks. The model features dual-mode functionality, allowing users to toggle between rapid responses and complex analysis, optimized for serverless deployment with a 99.9% SLA.

Trained on 630 billion tokens for 32K context and 209 billion for 128K, DeepSeek-V3.1 handles extended conversations and large-scale codebases with improved efficiency. Its thinking mode matches the quality of DeepSeek-R1 while significantly accelerating processing speeds for production environments.

The model's architecture supports diverse applications, though specific use cases remain undisclosed in this announcement. Its release underscores Together AI's focus on adaptable AI solutions for enterprise demands.