🚀 DeepSeek V3.1 Update Alert 🚀 Big wins in agent benchmarks and tool usage! 🎉 This hybrid model lets you toggle reasoning on and off. The instruct mode's performance has seen a noticeable bump, while reasoning mode is holding steady against R1-0528—bonus: it now uses fewer tokens with the same quality! 💡 We've also revamped the Base model for better long-context handling—10x context extension at 32k tokens and 3x at 128k tokens! 🙌 Plus, we trained in formats compatible with mxfp8 and mxfp4, which are now part of Blackwell. 🔄 API refresh: now includes Anthropic API compatibility for smooth Claude Code integration. Context length is boosted to 128k tokens, and we're testing new features for function calling! Happy hour pricing update starting September 5th! ⏰ Say goodbye to nightly discounts (China time) and pricing differences between reasoning and instruct modes. New output price: $1.68 per million tokens (current prices are $1.10 for instruct and $2.19 for reasoning). Input