Inkling 2 Mini hits 613 tok/s decode and 1M context on one RTX 5090 via Basalt

jesdga95 · reddit · 2026-11-21

Thinking Machines' small MoE model Inkling 2 Mini, which outperforms Qwen 3.8 27B on some tasks, was benchmarked by a Reddit user running it on a single RTX 5090 with Basalt 0.5.5.

Original post →

More from Infra

Infra channel →