Step 3.7 Flash – 198B-A11B MoE vision-language model | Dark Hacker News