Recommended default
GLM-5 / Kimi K2.6 / MiniMax-M2-class
Use FP8/INT8 where supported, or runtime-specific quantization. This is the highest-scoring current open-weight model that fits this card cleanly, selected by benchmark then fit then freshness, not by parameter count.
Benchmark anchor
MiniMax-M2 claims #1 open-source global composite by Artificial Analysis at release; verify against your own target benchmark.
Evidence
MiniMax-M2 release claims a #1 open-source composite; treat as a candidate and verify on your target benchmark. MI300X is positioned around memory capacity and runtime support.