AI BriefingMistral Benchmark on Strix Halo
·2026.05.04 03:49
Key point
On AMD Strix Halo, Mistral Medium 3.5 took about 2 hours to process a 48k prompt.
Details
On AMD Strix Halo, unsloth/Mistral-Medium-3.5-128B-GGUF:UD-Q5_K_XL was run via llama-server to handle codebase architecture questions. The settings were 80,000 context and reasoning_effort=high.
- Evaluating the 48,349-token prompt took 4,955,501ms, at a speed of 9.76 tok/s.
- 5,583 tokens generated took 2,652,690ms, with a generation speed of 2.10 tok/s.
- In practice, a task with a 48k prompt + 4k thinking tokens took about 2 hours, showing that while a 128B-class model can run on Strix Halo, the latency was very significant.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Step 5 Preview Released: 600B Parameter MoE Model Highlights Agentic Tasks and Financial Analysis Performance