Did anyone do a full bench of e.g. Qwen Flash Next IQ4 and Qwen 27b FP8? Here are some

I let Codex do some eval on Qwen 3.8 Flash Next IQ4_XS (served via vllm and r9v) and Qwen 3.8 FP8 (served via vllm and radiance). Here are the results: Benchmark Flash-Next IQ4_XS Qwen3.8 27B FP8 Result MMLU-Pro 83.8% 75.0% Flash-Next GPQA Diamond 42.5% 27.5% Flash-Next GSM8K 97.5% 90.0% Flash-Next