DeepSeek API enforces peak/off-peak pricing with up to 1100% surge
APPSO · wechat · 2026-08-18
Top story: DeepSeek API officially adopted peak/off-peak pricing from Aug 17. During peak hours (9:00–12:00, 14:00–18:00 Beijing time), DeepSeek-V4-Pro input rises from RMB 3 to 9 per million tokens and output from 6 to 27, with cached-hit input up 1100%; off-peak prices are roughly half, aiming to shift enterprise workloads away from congested hours. OpenRouter data shows China's weekly LLM token volume hit 342.5 trillion — first globally for 15 straight weeks — with DeepSeek-V4-Flash leading at 88.3 trillion.
Other highlights: OpenAI reportedly dissolved its Preparedness team in late July, its third frontier-risk team scrapped; Qwen3.8-27B passed 1M downloads in two days, topping HuggingFace trending; Alibaba's Qwen will power Apple Intelligence in China, with Apple and Alibaba also jointly training a dedicated model; Unitree lists on the STAR Market Aug 19 and demoed a "Superman" robot jumping 2m and sprinting at 12.66 m/s; OpenAI opened a 1M-token context option for GPT-5.6-Sol; Jensen Huang outlined plans to mobilize over $500B in third-party capital with Apollo, BlackRock, KKR and others for AI factories.
More from Models
- Qwen3.8-27B PrismaAqua benchmark: Near-BF16 quality — offgridai · 2026-08-18
- Harvard's Zak Kohane finds 5 AI detectors all flag his own writing as AI — zakkohane · 2026-08-18
- Sakana AI releases Japanese-specialized reasoning model Sakana Namazu — hardmaru · 2026-08-18
- Claim: DeepSeek V4 Beats Fable with J-Space Plugin Fixes — jmorant555 · 2026-08-18
- Running a fully local AI podcast station with Qwen — sysadmin420 · 2026-08-18
- Built a game in two prompts with Qwen 3.8 — lordekeen · 2026-08-18