Anthropic says Mythos got Claude running on AMD chips in 3 days, a job that used to take dozens of engineers months
According to a quote from Anthropic's IPO roadshow video shared on X, Anthropic started working with AMD GPUs for the first time this year and let its Mythos model try the bring-up. Anthropic says bringing on a new chip platform used to take dozens of engineers months, but Mythos got Claude running on AMD in about three days with basically no human intervention, porting or writing almost all of the GPU kernels. Coverage of the roadshow slides names AMD's MI355X and says the result was usable performance for prompt processing.
Why it matters
If models can port themselves to new hardware this fast, labs are less tied to a single chip vendor, which can affect price, capacity and how quickly new models reach users.
Who it's for
Developers and teams following AI infrastructure, GPU and CUDA-vs-AMD news, and Claude users curious about capacity.
What it means for you
- Treat the 3-day figure as Anthropic's own claim from marketing material for its IPO, not an independent benchmark.
- If you write GPU code, try asking Claude to port one small CUDA kernel to ROCm or HIP and check the output with tests.
- Watch for Anthropic's prospectus and AMD announcements for more detail on what ran and how fast.
Applies to: Applies to: Anthropic's internal infrastructure. Mythos 5.2 itself is limited to Anthropic's vetted verification programs, so this is not something you can run. Claim made in Anthropic's IPO roadshow materials; we read the quote and press coverage but did not watch the official video ourselves. Industry news, not investment advice. Checked on November 11, 2026.
Note: reported by users on X; we haven't independently confirmed it works in the current version.
Source: @AndrewCurran_ on X ↗
Written by AgentStackList from the original post. Checked Nov 11, 2026. Features, plans and availability can change, so confirm on the source before you rely on it.
anthropicmythosamdgpuindustry-news
More Claude tricks
- Stop reading tool-call logs: this free plugin shows a live map of everything Claude Code is doing · Nov 11
- Alibaba's new Qwen4-Max is live for just two weeks, at as little as 8% of the normal price, and it plugs into Claude Code · Nov 11
- Goldman took Claude Code away from bankers. JPMorgan just went the other way and gave it to non-engineers · Nov 11
- Claude Code read a 32-year-old Jeep's engine computer through a $3 Arduino, found why it kept failing smog, and the owner open-sourced the kit · Nov 11
Agent Stack Weekly · free
Get the week's best AI tricks
Limit and pricing changes, new MCP servers, and setup fixes for Claude Code, Codex, Cursor, Copilot and Antigravity.
No spam. Unsubscribe anytime. Read past issues