In a report published on 12 November, International Business Times described Anthropic's claim that its model helped get Claude running on AMD chips in three days, against work that previously required months. The task covered prompt processing for one model on AMD MI355X GPUs. That is the stage when a model processes input before generating a response.#
According to the reported investor presentation, Mythos wrote or adapted nearly all the necessary kernels, the software routines that execute calculations on chips. If the earlier work had comparable scope, automation could reduce the staff effort needed to use an alternative supplier. That would ease one barrier for companies adapting workloads to reduce dependence on a chip supplier.#
The result is limited. Running a workload at what the report calls a useful performance level does not establish production readiness or lower total operating costs. Prior engineering work or a narrower task could also explain the shorter implementation.#
Anthropic's clarification of the task boundaries could test the acceleration claim, though not production economics. Evidence that the three-day task excluded stages included in the earlier months-long work would undermine the interpretation that AI reduced comparable porting effort.#
