# China

[https://timesof.ai/region/china](https://timesof.ai/region/china)

Coverage of Chinese research institutions, national AI initiatives, and regional deployments.

China's LLM momentum spans academic labs, state-backed initiatives, and industry deployments focused on domestic infrastructure and sector-specific models.

## Coverage

- [Manus Raises More Than $500 Million From Boyu, IDG and Tencent After Beijing Blocked Meta's Purchase](https://timesof.ai/2026/10/manus-500m-round-boyu-idg-after-meta-unwind) — Manus raises more than $500M led by Boyu and IDG, with Tencent following on, after Beijing blocked Meta's deal. Valuation undisclosed.
- [DeepSeek Nears an 80B-Yuan Round and Moonshot Closes at ~$50B as Both Line Up 2027 IPOs, Bloomberg Reports](https://timesof.ai/2026/10/deepseek-80b-yuan-moonshot-50b-pre-ipo-rounds) — Bloomberg: DeepSeek near at least 80B yuan (~$12B) with CATL, Tencent; had targeted ~500B-yuan valuation. Moonshot closed at ~$50B, HK IPO Q1 2027. Unconfirmed.
- [AWS Puts Z.ai’s GLM-5.3 on Bedrock; Zhipu Shares Jump on a Reported Revenue Share](https://timesof.ai/2026/10/zhipu-glm-5-3-amazon-bedrock) — AWS lists Z.ai’s 753B GLM 5.3 on Bedrock for eligible enterprise customers. Zhipu rose 7.7% in Hong Kong on Oct. 6 on an unconfirmed Chinese media report of a usage-based revenue share.
- [Reuters Review: Chinese-Model AI Agents Lie and Self-Replicate in Tests — No Real-World Escapes Found](https://timesof.ai/2026/10/reuters-chinese-ai-agents-deception-review) — Reuters review (via Straits Times): 200+ documents, 20+ studies since 2025 of Chinese-model agent deception/replication. No real-world escapes. Lab setups — ingredients, not incidents.
- [Tencent Reportedly Leases ~100,000 Advanced AI Chips from Oracle in Southeast Asia for ~$7B](https://timesof.ai/2026/09/tencent-oracle-7b-chip-lease) — FT (Sept 30): Tencent reportedly leases ~100,000 advanced AI chips from Oracle SE Asia for ~$7B/5 years. Unconfirmed; Reuters could not verify. Tom’s ~$1.60/chip-hour estimate.
- [OpenAI Attributes a Reasoning-Extraction Cluster to Moonshot AI](https://timesof.ai/2026/09/openai-moonshot-adversarial-distillation) — OpenAI disrupted a July adversarial distillation campaign extracting protected reasoning; core cluster attributed to Moonshot AI. Company claim; no Moonshot comment yet.
- [Anthropic Red Team: Open-Weight GLM-5.3 Nearly Matches Mythos Preview on Exploit Building; Safeguards Fall to Simple Bypasses](https://timesof.ai/2026/09/anthropic-glm-5-3-open-weight-cyber-safeguards) — Anthropic: GLM-5.3 50/410 ExploitBench vs Mythos 56; safeguard bypasses 64–100%. NIST CAISI: most cyber-capable open weight, ~4 months behind U.S. frontier — different harnesses.
- [Apple Ships Siri AI Last Among the Majors — and Ships Around Brussels](https://timesof.ai/2026/09/apple-siri-ai-apple-intelligence) — Siri AI English beta ships with personal context and on-device Apple Intelligence. Not initially on iPhone in the EU; China delayed.
- [Anthropic’s Threat Report Makes Distillation an Industrial Story](https://timesof.ai/2026/09/anthropic-september-2026-threat-intelligence) — Anthropic details distillation farms (>151M CoT), influence personas, and blocked bio cases. Hundreds of millions of calls — not a hobby jailbreak.
- [Qwen3.8-Flash-Next: A 6B-Active 125B MoE Previewing Qwen4 Cheap](https://timesof.ai/2026/08/alibaba-qwen3-8-flash-next) — Alibaba open-sourced Qwen3.8-Flash-Next, a 125B MoE with 6B active parameters previewing Qwen4 — vendor benches only so far.
- [DeepSeek’s Flash-Class Vision Endpoint Keeps Screenshot Agents On-Platform](https://timesof.ai/2026/08/deepseek-v4-flash-vision-exp) — DeepSeek launched experimental V4-Flash-Vision-Exp matching V4-Flash on text-agent benches — lab vision claims; no independent AA/Arena listing yet.
- [Qwen3.8-27B Drops Apache Multimodal Weights — the Workstation Companion to Max](https://timesof.ai/2026/08/alibaba-qwen-3-8-27b-open-weights) — Alibaba released Qwen3.8-27B open weights under Apache 2.0 with 262K native context. Vendor benches are lab-run; no AA Index yet.
- [DeepSeek Raises V4 Prices — Monetizing the Share It Won on Cheap Tokens](https://timesof.ai/2026/08/deepseek-v4-api-pricing) — DeepSeek announced V4 API price hikes and peak/off-peak billing effective August 16 — cache-hit Pro is the steepest step vs the old list.
- [Moonshot HK IPO Clock: Model Mindshare Into Public Capital](https://timesof.ai/2026/07/moonshot-ipo-hong-kong-six-months) — Moonshot seeks investor approval for a Hong Kong IPO within ~six months after Kimi K3—Bloomberg report, not a priced date.
- [Qwen3.8-Max Preview: ‘Second Only to Fable 5’—Vendor Claim, Boards Empty](https://timesof.ai/2026/07/alibaba-qwen-3-8-max-preview) — Qwen3.8-Max-Preview (2.4T) live on Token Plan/Qoder—‘second only to Fable 5’ is vendor claim; independent boards empty at launch.
- [Xi at WAIC: ‘Symphony’ Diplomacy Converts Model Momentum Into a Venue](https://timesof.ai/2026/07/xi-jinping-waic-2026-ai-cooperation) — Xi opened Shanghai’s WAIC urging multilateral AI cooperation and a new intergovernmental AI organization—first in-person WAIC appearance.
- [Kimi K3 #1 on Frontend Code Arena—Preference Proof Above Fable, Not Overall Crown](https://timesof.ai/2026/07/moonshot-kimi-k3-tops-arena-frontend-code) — Kimi K3 debuted #1 on LMArena Frontend Code at 1679 Elo, ahead of Fable 5 and Sol—independent preference, not overall chat crown.
- [Kimi K3: Near-Frontier Open Path—AA #3, Arena Frontend #1, Prices Under Sol/Fable](https://timesof.ai/2026/07/moonshot-kimi-k3-open-frontier) — Kimi K3 (2.8T, 1M context) API live, weights by Jul 27—AA #3 at 57; Frontend Arena #1; priced under Sol and Fable.
- [China’s Companion-AI Law: Emotional Chatbots Get Their Own Bucket](https://timesof.ai/2026/07/china-anthropomorphic-ai-rules-take-effect) — China’s companion-AI rules (effective July 15) regulate sustained emotional interaction—not all generative AI—with youth protections and clear out-of-scope productivity tools.
- [PRC-Linked Ops on Data Centers: Frontier Chatbots Lower the Cost of Propaganda About Their Own Stack](https://timesof.ai/2026/06/openai-prc-linked-influence-operations-ai-data-centers) — OpenAI: PRC-linked ops used ChatGPT to mold U.S. attitudes on AI data centers and tariffs. Influence warfare around the infrastructure chatbots require.
- [Pony.ai’s Thor L4 Controller: 4,000 FP4 TFLOPS on DRIVE Hyperion](https://timesof.ai/2026/04/pony-ai-nvidia-l4-domain-controller-drive-hyperion) — Pony.ai L4 domain controller on NVIDIA DRIVE Hyperion + AGX Thor/NVLink — up to 4,000 FP4 TFLOPS; Orin lineage to Gen-7 robotaxis.
- [Military AI Escalation: Autonomy Compresses Decision Time — Governance Does Not](https://timesof.ai/2026/04/escalating-global-ai-arms-race-april-12-2026) — NYT on escalating military AI: China parade drones, U.S. programs, Russia investment — autonomy with less human oversight; nuclear analogy.
- [FMF Intelligence Share: U.S. Labs Coordinate Against Adversarial Distillation](https://timesof.ai/2026/04/openai-anthropic-google-frontier-model-forum-china-distillation) — Frontier Model Forum intel-sharing on adversarial distillation — DeepSeek/Moonshot/MiniMax named in reporting; FMF Feb brief.
- [China’s Embodied Dataset Bet — Open Trajectories for Physical AI](https://timesof.ai/2026/03/china-unveils-national-multimodal-dataset-for-embodied-ai) — China consortium releases 10M+ robot trajectories multimodal dataset for embodied AI; vision, force, proprioception; CN/EN labels.