London-based AI task-to-model router Callosum reportedly raises $100M seed round

Callosum provides a routing layer that sits between user prompts and inference endpoints. The software profiles each incoming task, then dispatches it to the model and accelerator that minimise cost or latency. This removes the need for engineers to hard-code model selection or maintain multiple API integrations. The trade-off is a dependency on Callosum’s routing logic and pricing model, which may change after the seed capital is spent.

The $100M seed round is an outlier for early-stage infrastructure plays. It suggests that investors view automated model selection as a critical bottleneck in AI deployment. Teams building their own routing logic can now benchmark against Callosum’s commercial offering, but must weigh the risk of vendor lock-in against the speed of integration. The round size also implies that Callosum will need to scale quickly to justify the valuation.

Callosum’s approach stops working when workloads require custom model fine-tuning or proprietary data that cannot be exposed to a third-party router. It also assumes that model performance is stable enough for offline profiling; real-time drift in model quality or pricing could degrade routing accuracy. Engineers evaluating Callosum should test its behaviour under sudden traffic spikes or model deprecations to assess resilience.