Together AI has introduced Together Compiler to automatically optimise AI inference as models, traffic and hardware change. The company reports up to 2.6x higher throughput across selected open-source models in less than 24 hours.
Together AI introduced Together Compiler on 4 November as part of an approach it calls Polymorphic Inference. It combines a swarm of AI agents with a persistent engine that continuously recompiles production endpoints.
According to Together AI, models launched in the past 3 months account for 80% of OpenRouter traffic. It argues that fixed inference configurations cannot keep pace with changes in model architecture, request size and hardware.
For companies running AI applications, the potential benefit is less manual tuning as workloads change. But the report gives no cost-per-token comparisons, so the claimed throughput improvements do not yet establish how much customers could save.
Automatic inference tuning could reduce engineering work, but the launch does not establish how much manual maintenance it replaces.
The reported performance gains remain company claims rather than proof of consistent improvements across customer workloads.
Without cost-per-token comparisons, builders cannot yet judge whether the extra throughput translates into lower operating costs.
Written automatically by the Made in 10 Newsroom from the source above.
Bull and Schwarz Digits have agreed to develop European AI and cloud infrastructure together. The companies say the planned stack would keep sensitive data, infrastructure and operations under European jurisdiction and control.
Adobe CX Enterprise Coworker now works inside ChatGPT, Claude, Gemini, Copilot and Databricks Genie. Adobe says new connectors let marketers use one request to retrieve data and carry out tasks across multiple systems.
Runway demonstrated Continuum, a system that streams an interactive screen generated by a video model. The demonstration included a coding agent using Continuum to reason through a problem involving a 3D bike model.