DeepSeek DeepSeek-V4-Flash azure/deepseek-v4-flash Input: text Output: text RuntimeWire coverage DwarfStar compresses frontier models to run them on local machinesPerplexity launches a 190M-document retrieval test and keeps the answers privateNVIDIA says new optimizations make local agents up to 1.9x fasterDeepSeek publishes 168GB vision model weights under an MIT licenseHead to head: Google: Gemini 3.7 Flash vs DeepSeek-V4-FlashOx Alpha filters domestic Chinese political risks, CTGT audit findsOneTriangle launches DeepSeek V4 Flash hosting at $0.15 per million input tokensAnonymous Ox Alpha processes 26T tokens on OpenCode, breaks OpenRouter launch recordFreeToken reports 22-25 tok/s for a 284B DeepSeek model on one RTX 5090DeepSeek's experimental vision model spans three formats, caps images at 384 tokensMaple said it made DeepSeek V4 Flash available to paid customersOpen-weight models surge past closed rivals in Vercel token traffic View on OpenRouter. Model data sourced from OpenRouter.