Benchmark
MMLU-ProX
A multilingual extension of MMLU-Pro that evaluates the comprehensive knowledge capabilities of large models across multilingual, multi-task scenarios.
Timeline (last 90 days)
Oct 16
Comparing BF16 and NVFP4 under the same harness, WebVoyager dropped from 77.72 to 66.31, DeepSWE from 24.65 to 18.2, and MMLU-ProX from 76.91 to 73.76, with other rows within the noise range.
Unverified50%