Unverified50% confidenceFactExact time
A user runs the 397B model in FP8 on 8× GB10 units at a speed 20% faster than INT4, and published the first solution for MiMo 2.5 Pro and Kimi 2.6 on the NVIDIA forums.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/4/2026
First Seen
Valid until: 1/2/2027
Sources
Related Entities
Related Claims
UnverifiedMiMo V2.6 Flash has 310B total parameters and 15B activated parameters, with a speed nearly twice that of Pro.68% similarUnverifiedThe MI355X's memory capacity exceeds that of the NVIDIA H100 (80 GB) and H200 (141 GB).67% similarUnverifiedMiMo Pro is a 1-trillion-parameter, 2-billion-active-parameter MoE model that requires approximately 500 GB of VRAM—about 8 H100s—to run at 4-bit quantization.67% similarUnverifiedCompared to the previous-generation MI300 series, the MI355X has improvements in memory bandwidth, compute density, and energy efficiency.66% similarUnverifiedMiMo-V2.6 supports a context length of up to 1 million (1M) tokens.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/974694API
curl https://kongchang.com/api/v1/knowledge/claims/974694MCP
get_claim(id=974694)