The Number
- 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.
- Alibaba's qwen3.8-max ranks #6 overall at 1497 Elo on Arena's text preference ranking, the top-placed China lab, per LMArena as of 2026-08-09.
- DeepSeek lists deepseek-v4-pro at $0.435/M input on a cache miss with $0.87/M output, per its pricing docs as of 2026-08-10.
- Qwen's cheapest hosted tier, qwen3.7-flash at $0.03/M input, undercuts every China-lab listing on the platform, on OpenRouter as of 2026-08-10.
Top of the Fold
- Moonshot lists Kimi K3 as its flagship with a 1M-token context on its API pricing page, retiring the Moonshot V1 series platform-wide by 31 August, per platform.moonshot.cn as of 2026-08-10.
- DeepSeek warns of a significant near-term price increase across its API in its own pricing docs, a reversal of the cheap-token posture that built its developer base, per api-docs.deepseek.com as of 2026-08-10.
- Moore Threads posts 1.736B yuan in first-half revenue, already past its full 2025 total and up 147% year on year, per its interim report filed 2026-08-09.
Untranslated
- A Chinese team says it ran an end-to-end data-layer RSI loop, having a model generate its own training problems as public benchmarks run dry, per 量子位 (verify claim independently).
- Moonshot's docs confirm Kimi K2.7 Code now runs as a multimodal coding model alongside the new K3 flagship, per platform.moonshot.cn as of 2026-08-10.
- Otherwise quiet across the Chinese community overnight.
The Kicker
- The Beijing 百花奖 added a short-drama honors slate for the first time, handing five annual awards to 15 titles including 《食神2026》, per 雷峰网 on 2026-08-09.
