The Overnight — GeekPark https://about.geekpark.net/overnight/ What China's AI world did while the West was asleep — daily China-AI briefing from GeekPark. en The Overnight https://about.geekpark.net/images/og-image.png https://about.geekpark.net/overnight/ Unitree lands on STAR Market as A-share's first humanoid stock https://about.geekpark.net/overnight/2026-08-19/ DeepSeek-V4-Pro (0813) runs $0.66/M input at cache-miss off-peak, cheaper than any US flagship listing, per DeepSeek's pricing docs as of 2026-08-19. Alibaba's Wed, 19 Aug 2026 08:02:23 GMT https://about.geekpark.net/overnight/2026-08-19/ DeepSeek's new peak-off-peak pricing goes live; Alibaba drops an AI music model https://about.geekpark.net/overnight/2026-08-18/ DeepSeek-V4-Pro now runs $0.66/M cache-miss input off-peak, half its $1.32 peak rate under first-ever peak/off-peak tiers, per DeepSeek's pricing docs (as of Tue, 18 Aug 2026 08:02:08 GMT https://about.geekpark.net/overnight/2026-08-18/ DeepSeek V4's new peak-and-off-peak prices go live at midnight https://about.geekpark.net/overnight/2026-08-17/ DeepSeek V4-Pro cache-miss input runs $0.66/M off-peak against $1.32 at peak, new tiered rates as of 2026-08-17, per its pricing docs. Qwen3.8-max sits at 8 on Mon, 17 Aug 2026 08:02:02 GMT https://about.geekpark.net/overnight/2026-08-17/ DeepSeek splits its API into peak and off-peak pricing https://about.geekpark.net/overnight/2026-08-16/ DeepSeek-V4-Pro cache-miss input runs $0.435/M off-peak versus $0.87/M output, before a peak/off-peak split takes effect 16:00 UTC on Aug 16, per DeepSeek's Sun, 16 Aug 2026 08:02:31 GMT https://about.geekpark.net/overnight/2026-08-16/ Qwen3.8-27B goes open on consumer GPUs; GLM-5.3 and DeepSeek Harness crowd the coding lane https://about.geekpark.net/overnight/2026-08-15/ DeepSeek-V4-Pro-0813 cache-miss input runs $0.435/M against $0.87/M output, before a peak/off-peak split takes effect 16:00 UTC Aug 16, per DeepSeek's pricing Sat, 15 Aug 2026 08:02:45 GMT https://about.geekpark.net/overnight/2026-08-15/ DeepSeek ships Harness, and lifts V4-Pro prices in the same night https://about.geekpark.net/overnight/2026-08-14/ DeepSeek-V4-Pro (0813) lists at $0.435/M input (cache miss) and $0.87/M output, ahead of the August 16 shift to peak/off-peak billing, per DeepSeek's pricing Fri, 14 Aug 2026 08:03:19 GMT https://about.geekpark.net/overnight/2026-08-14/ DeepSeek V4 Pro ships, Manus splits from Meta, Lin Junyang's startup surfaces https://about.geekpark.net/overnight/2026-08-13/ DeepSeek V4 Pro lists at $0.435/M input (cache miss), $0.87/M output, with docs flagging a coming price rise, per DeepSeek's pricing page 2026-08-13. Thu, 13 Aug 2026 08:02:24 GMT https://about.geekpark.net/overnight/2026-08-13/ Ant funds robot fingertips; Kimi K3 tops the China code board https://about.geekpark.net/overnight/2026-08-12/ DeepSeek-V4-Pro sits at $0.435/M input cache-miss, still under a tenth of most US flagships, per DeepSeek's pricing docs as of 2026-08-12. Kimi-K3-Max holds 2 Wed, 12 Aug 2026 08:01:59 GMT https://about.geekpark.net/overnight/2026-08-12/ Qwen App goes paid, Kimi K3 lands #2 on Arena coding https://about.geekpark.net/overnight/2026-08-11/ Kimi K3-max sits 2 on Arena's coding preference ranking at Elo 1675, the top China entry, behind only Claude Opus 5-max, per LMArena as of 2026-08-10. Tue, 11 Aug 2026 08:03:41 GMT https://about.geekpark.net/overnight/2026-08-11/ Kimi K3 Max holds #2 on Arena's coding board as DeepSeek warns of a price hike https://about.geekpark.net/overnight/2026-08-10/ Moonshot's kimi-k3-max sits 2 on Arena's coding preference ranking at 1675 Elo, behind only Anthropic's claude-opus-5-max, per LMArena as of 2026-08-09. Mon, 10 Aug 2026 08:03:17 GMT https://about.geekpark.net/overnight/2026-08-10/ Apple ships Qwen into China Macs, then pulls the page https://about.geekpark.net/overnight/2026-08-09/ DeepSeek-V4-Flash sits at $0.14/M input on a cache miss, with cache-hit input at $0.0028/M, per its pricing docs as of 2026-08-09. Qwen3.7-Max runs a 50% promo Sun, 09 Aug 2026 08:03:34 GMT https://about.geekpark.net/overnight/2026-08-09/ ByteDance's trillion-parameter model, and Kimi K3 climbs Arena code https://about.geekpark.net/overnight/2026-08-08/ DeepSeek-V4-Flash lists a cache-miss input price of $0.14/M tokens against $0.435 for V4-Pro, with a warning that overall pricing rises soon, per DeepSeek's Sat, 08 Aug 2026 08:01:56 GMT https://about.geekpark.net/overnight/2026-08-08/ DeepSeek warns of a big price hike as Meta undercuts it on cost https://about.geekpark.net/overnight/2026-08-07/ DeepSeek-V4-Flash lists at $0.14/M cache-miss input and $0.28/M output as of Aug 7, with the docs warning of a coming increase, per its pricing page. Fri, 07 Aug 2026 08:02:03 GMT https://about.geekpark.net/overnight/2026-08-07/ Sand.ai open-sources the first 100B-plus MoE video model https://about.geekpark.net/overnight/2026-08-06/ Qwen3.8-max sits 5 overall on Arena's text preference ranking at Elo 1496, the top China-lab entry, as of 2026-08-05. Kimi K3-max ranks 2 on Arena's coding Thu, 06 Aug 2026 08:01:47 GMT https://about.geekpark.net/overnight/2026-08-06/ Qwen3.8-Max lands #4 on Arena coding as overseas platforms scramble to add it https://about.geekpark.net/overnight/2026-08-05/ Alibaba's qwen3.8-max sits 4 on Arena's coding preference board at Elo 1668, top China entry behind Anthropic's Opus, as of Aug 4, per LMArena. Moonshot's Wed, 05 Aug 2026 08:01:41 GMT https://about.geekpark.net/overnight/2026-08-05/ Alibaba ships Qwen3.8-Max at 2.4T params, three giants converge on AI office https://about.geekpark.net/overnight/2026-08-04/ Qwen3.8-Max lists at 12 yuan per 1M input tokens (36 output) in China, 1.5 yuan on implicit cache hit, per Alibaba's DashScope pricing (2026-08-04). Tue, 04 Aug 2026 08:02:03 GMT https://about.geekpark.net/overnight/2026-08-04/ Kimi K3 Max cracks Arena's coding top two; two Chinese video models drop same day https://about.geekpark.net/overnight/2026-08-03/ Moonshot's kimi-k3-max sits 2 on Arena's coding preference ranking at 1676 Elo, the only China entry inside the top five as of Aug 2, per LMArena. Z.ai's Mon, 03 Aug 2026 08:01:35 GMT https://about.geekpark.net/overnight/2026-08-03/ Kimi K3 Max holds #2 on Arena's coding board; DeepSeek V4-Flash goes public beta https://about.geekpark.net/overnight/2026-08-02/ Moonshot's kimi-k3-max sits 2 on Arena's coding preference ranking at Elo 1675, the top China entry and behind only Anthropic's claude-opus-5-max, per LMArena Sun, 02 Aug 2026 08:01:43 GMT https://about.geekpark.net/overnight/2026-08-02/ Kimi K3 Max cracks Arena's top code tier; DeepSeek V4-Flash ships final https://about.geekpark.net/overnight/2026-08-01/ Moonshot's kimi-k3-max sits 2 on Arena's coding preference ranking at 1682 Elo, the only non-Anthropic model in the top four, per LMArena as of 2026-07-31. Sat, 01 Aug 2026 08:01:48 GMT https://about.geekpark.net/overnight/2026-08-01/ Moonshot's Kimi K3 ships, and Chinese money keeps chasing it https://about.geekpark.net/overnight/2026-07-31/ What China's AI world did while the West was asleep. Fri, 31 Jul 2026 08:01:51 GMT https://about.geekpark.net/overnight/2026-07-31/ Kimi K3 Max lands #2 on Arena's coding board; Kimi closes $3.5B https://about.geekpark.net/overnight/2026-07-30/ What China's AI world did while the West was asleep. Thu, 30 Jul 2026 11:22:35 GMT https://about.geekpark.net/overnight/2026-07-30/ DeepSeek halts second funding round; Kimi K3 lands on Arena https://about.geekpark.net/overnight/2026-07-28/ What China's AI world did while the West was asleep. Tue, 28 Jul 2026 13:34:57 GMT https://about.geekpark.net/overnight/2026-07-28/