Anthropic says Claude Mythos Preview cut AMD MI355X onboarding from up to 12 months to around three days

What happened
Anthropic used its Claude Mythos Preview AI model to automate the process of adapting its software to run on AMD MI355X GPUs. This task, which previously required six to twelve months and a large engineering team, was completed in about three days with minimal human involvement. The AI handled nearly all of the low-level code needed for the port. Anthropic shared these results with IPO investors, highlighting the efficiency of the new approach.
Key details
- Claude Mythos Preview ported Anthropic's model to AMD MI355X GPUs in about three days.
- The previous process took six to twelve months and required dozens of engineers.
- The AI wrote almost all of the necessary GPU kernels for the port.
- Anthropic manages its models across Nvidia, AMD, Google TPU, and AWS Trainium hardware.
- The company treats its compute resources as a single interchangeable pool.
Why it matters
Companies developing AI models can now move their software to new types of GPUs much faster and with fewer engineers, reducing both time and labor costs. This change affects hardware suppliers like Nvidia, whose CUDA software has made switching difficult in the past, and opens opportunities for competitors like AMD. Organizations running large AI workloads gain more flexibility in choosing hardware, which can impact purchasing decisions and infrastructure planning.
