The Definitive Guide toAI Data Centers
Ask the GuideAboutAccount
Guide › Numbers register › Claim

NVIDIA DeepSeek-R1-0528 comparison: MMLU-Pro 85% FP8 versus 84% NVFP4, a one-percentage-point decrease

1 percentage pointderived

Value kindderived — Derived values are guide calculations; their result is only as fixed as the stated inputs, scope, and method.
ScopePublished rounded scores for one model and recipe, not a universal NVFP4 accuracy guarantee.
As of2025-06
SourceNVIDIA — Figure 6, MMLU-Pro rounded scores 85 and 84. · Figure 6, MMLU-Pro rounded scores 85 and 84
Derivation85% − 84% = 1 percentage point; relative decline = 1/85 = 1.176%.
Reviewchecking…review by 2026-12-05 · standard cadence
Recorded changeslast 2026-09-16 · 2 revisions tracked
Claim idnvfp4-inference-accuracy-drop-vs-fp8-on

Where the guide uses it

← Full numbers register — every date-stamped figure in the guide, with revision history.