DeepSeek on X: "🚀 Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient. 🔹 Introducing the smallest model in our new architecture family, with native visual understanding. 🔹 Designed for greater capability, faster inference, higher throughput, and scaling to larger models. 1/6" / X (../../../x.com/favicon.ico) (https://x.com/apple-touch-icon.png) (../../../x.com/manifest.json) (https://abs.twimg.com) (../../../abs.twimg.com/fonts/subset/Chirp-Regular.c88864db.latin.woff2) (../../../abs.twimg.com/fonts/subset/Chirp-Medium.ca6d679a.latin.woff2) (../../../abs.twimg.com/fonts/subset/Chirp-Bold.d8ac01b2.latin.woff2) (https://api.x.com) (https://pbs.twimg.com) (https://t.co) (https://video.twimg.com) (https://ton.twimg.com) (https://cdn.syndication.twimg.com) (https://x.com/os-x.xml) (X) (https://x.com/os-grok.xml) (Grok) (https://x.com/sitemap.xml) (2097930608790167907.html) (https://publish.x.com/oembed?url=https%3A%2F%2Fx.com%2Fdeepseek_ai%2Fstatus%2F2097930608790167907) (../../../pbs.twimg.com/media/HR1UyHiaAAAtpqw@format=webp&name=medium) (../../../abs.twimg.com/x-web/x-web/assets/styles-D__51vkT.css) (https://x.com/) Post (https://x.com/i/jf/onboarding/web?mode=login&redirect_after_login=%2Fdeepseek_ai%2Fstatus%2F2097930608790167907) Log in (https://x.com/i/jf/onboarding/web?mode=signup&redirect_after_login=%2Fdeepseek_ai%2Fstatus%2F2097930608790167907) Sign up Post DeepSeek on X: "🚀 Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient. 🔹 Introducing the smallest model in our new architecture family, with native visual understanding. 🔹 Designed for greater capability, faster inference, higher throughput, and scaling to larger models. 1/6" (https://x.com/deepseek_ai) (@deepseek_ai) (https://x.com/deepseek_ai) DeepSeek (https://x.com/deepseek_ai) @deepseek_ai 🚀 Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient. 🔹 Introducing the smallest model in our new architecture family, with native visual understanding. 🔹 Designed for greater capability, faster inference, higher throughput, and scaling to larger models. 1/6 (https://x.com/deepseek_ai/status/2097930608790167907/photo/1) () (https://x.com/deepseek_ai/status/2097930608790167907/photo/1) (2097930608790167907.html) 6:10 AM · Sep 10, 2026 (2097930608790167907.html) 933.1K Views (https://x.com/i/status/2097930608790167907) 463 1.4K 12K 1.7K Log in or sign up for X See what’s happening and join the conversation (https://x.com/i/jf/onboarding/web?mode=signup&redirect_after_login=%2Fdeepseek_ai%2Fstatus%2F2097930608790167907) Continue with phone Continue with Apple Continue with Google or (https://x.com/i/jf/onboarding/web?mode=login&redirect_after_login=%2Fdeepseek_ai%2Fstatus%2F2097930608790167907) Log in with username or email Relevant people (https://x.com/deepseek_ai) (Avatar) (https://x.com/deepseek_ai) DeepSeek @deepseek_ai (https://x.com/i/jf/onboarding/web?mode=login&redirect_after_login=%2Fdeepseek_ai) Follow Trending now (https://x.com/tos) Terms · (https://x.com/privacy) Privacy · (https://support.x.com/articles/20170514) Cookies · (https://help.x.com/resources/accessibility) Accessibility · (https://help.x.com/en/rules-and-policies/us-tida) US TIDA · (https://business.x.com/en/help/troubleshooting/how-twitter-ads-work.html?ref=web-twc-ao-gbl-adsinfo&utm_source=twc&utm_medium=web&utm_campaign=ao&utm_content=adsinfo) Ads Info · More © 2026 X Corp. (https://x.com/deepseek_ai) (@deepseek_ai) (https://x.com/deepseek_ai) DeepSeek (https://x.com/deepseek_ai) @deepseek_ai 🚀 Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient. 🔹 Introducing the smallest model in our new architecture family, with native visual understanding. 🔹 Designed for greater capability, faster inference, higher throughput, and scaling to larger models. 1/6 (https://x.com/deepseek_ai/status/2097930608790167907/photo/1) () (https://x.com/deepseek_ai/status/2097930608790167907/photo/1) (2097930608790167907.html) 6:10 AM · Sep 10, 2026 (2097930608790167907.html) 933.1K Views (https://x.com/i/status/2097930608790167907) 463 1.4K 12K 1.6K (https://x.com/deepseek_ai) (@deepseek_ai) (https://x.com/deepseek_ai) DeepSeek (https://x.com/deepseek_ai) @deepseek_ai (https://x.com/deepseek_ai/status/2097930613101838709) 3h 🧠 Asymmetric architecture. More intelligence, less cost. 🔹 552B-parameter MoE. 🔹 New Causal Encoder–Decoder architecture: just 8B active parameters for input, 16B for output. 🔹 New pre-training methods + larger-scale RL post-training deliver benchmark results ahead of Show more (https://x.com/deepseek_ai/status/2097930613101838709/photo/1) () (https://x.com/deepseek_ai/status/2097930613101838709/photo/1) (https://x.com/i/status/2097930613101838709) 29 110 1.4K 142K (https://x.com/deepseek_ai) (@deepseek_ai) (https://x.com/deepseek_ai) DeepSeek (https://x.com/deepseek_ai) @deepseek_ai (https://x.com/deepseek_ai/status/2097930617396887773) 3h 💾 Smaller KV cache. Bigger savings. Compared with the previous generation, V4.1-Flash’s KV cache needs just: 🔹 1/4 the HBM 🔹 1/8 the SSD storage Cache-hit charges often account for a large share of agent costs. Compressing the cache cuts those costs significantly. 3/6 (https://x.com/deepseek_ai/status/2097930617396887773/photo/1) () (https://x.com/deepseek_ai/status/2097930617396887773/photo/1) (https://x.com/i/status/2097930617396887773) 11 60 1.2K 92K (https://x.com/deepseek_ai) (@deepseek_ai) (https://x.com/deepseek_ai) DeepSeek (https://x.com/deepseek_ai) @deepseek_ai (https://x.com/deepseek_ai/status/2097930620680941732) 3h ⚡ V4.1-Flash is now live on the DeepSeek API with native multimodal support. Set your model to deepseek-flash. 🔹 V4-Flash & V4-Flash-Vision-Exp are retired. For compatibility, deepseek-v4-flash and deepseek-v4-flash-vision-exp temporarily route to V4.1-Flash. 🔹 Tests by Show more (https://x.com/i/status/2097930620680941732) 10 26 812 99K (https://x.com/deepseek_ai) (@deepseek_ai) (https://x.com/deepseek_ai) DeepSeek (https://x.com/deepseek_ai) @deepseek_ai (https://x.com/deepseek_ai/status/2097930624598446137) 3h 💰 More efficient architecture. Lower API prices. V4.1-Flash lets us serve more users at a lower cost. We’re passing the savings on to you. 🔹 Peak/off-peak pricing continues to balance demand. 🔹 Off-peak rates are 50% of peak rates. Schedule flexible workloads off-peak to Show more (https://x.com/deepseek_ai/status/2097930624598446137/photo/1) () (https://x.com/deepseek_ai/status/2097930624598446137/photo/1) (https://x.com/i/status/2097930624598446137) 11 25 799 70K (https://x.com/deepseek_ai) (@deepseek_ai) (https://x.com/deepseek_ai) DeepSeek (https://x.com/deepseek_ai) @deepseek_ai (https://x.com/deepseek_ai/status/2097930627949711516) 3h 🌐 Supporting open source. Expanding deployment options. We’ll work closely with the open-source community on V4.1-Flash inference support and explore more deployment options. Planning a large-scale deployment with 2,000 GPUs + a storage cluster? Let’s talk. 🔹 Model: Show more (https://t.co/rv2G2TN66a) () deepseek-ai/DeepSeek-V4.1-Flash · Hugging Face (https://t.co/rv2G2TN66a) From huggingface.co (https://x.com/i/status/2097930627949711516) 34 106 1.2K 92K (https://x.com/UnslothAI) (@UnslothAI) (https://x.com/UnslothAI) Unsloth AI (https://x.com/UnslothAI) @UnslothAI (https://x.com/UnslothAI/status/2097947144351252497) 2h Congrats DeepSeek on another epic release! Hopefully you guys will release smaller models for people to run locally. 🙏🐋 It's great that DeepSeek-V4.1 has 196B engram making it more accessible. (https://x.com/i/status/2097947144351252497) 8 9 336 17K