2026-10-13
Post reports Nvidia's China AI-chip share fell to 34% in H1 2026 as Huawei climbed to 28%
Huawei shipped 572K accelerators, and Jensen Huang is quoted blaming US export controls for handing over a $50B market
Story archive
2026-10-13
Anthropic red team warns small AI models could self-replicate on consumer devices within months; OpenAI pledges $1B for cybersecurity
Reports detail GLM-5.2 as only open-weight model completing full replication cycle, while DeepSeek plans to lead Winverse AI seed round
2026-10-13
Anthropic reports 16M+ Claude exchanges via 24,000 fraudulent accounts; names DeepSeek, Moonshot, MiniMax in distillation crackdown
Posts detail Zhipu and Moonshot campaigns, including 151M Alibaba-linked conversations and 300K Kimi requests rerouted to Claude
2026-10-13
DeepSeek reportedly in talks for up to $12B+ funding round, mulls robotics push
Reports cite Tencent and CATL as potential backers; separate reports say DeepSeek is leading a seed round in robotics startup Winverse AI.
2026-10-13
Qwen Image 2.1 Pro and Turbo models announced with 2K generation and fast 8-step output
Post by icreat_ai reports the new image models are live, with Turbo capable of producing an image in about 3 seconds in a real test.
2026-10-13
Muse Wonder launch announced as closed-source frontier model family, breaking from Llama's open-weight legacy
Post reports Wonder Full scores 56 on Artificial Analysis, matching Claude Sonnet 5.5 and exceeding GPT-6 Astra's 53, with $4/$20 API pricing and 1M token context
2026-10-13
Muse Wonder launches as closed-source frontier model family, breaking from Llama's open-weight tradition
Post reports Wonder Full scores 56 on Artificial Analysis, matching Claude Sonnet 5.5; API pricing set at $4/$20 with 1M token context
2026-10-13
DeepSeek reportedly nears $12B+ funding round, Tencent and CATL among backers
Reports also cite possible 2027 IPO and interest in embodied AI
2026-10-13
DeepSeek reportedly closing ~$12B round with Tencent and CATL, eyeing Hong Kong IPO
Reports vary on size and listing venue, with one account citing a possible domestic IPO and a valuation around $74B
2026-10-13
DeepSeek reportedly to lead seed round for robotics startup Winverse AI
Bloomberg-sourced reports say the deal values the months-old Hangzhou firm at about $400 million post-money, with Meituan's Long-Z also participating.
2026-10-13
DeepSeek reportedly to lead seed round for robotics startup Winverse AI
Bloomberg-sourced posts say the deal values the Hangzhou firm at about $400 million post-money; not final
2026-10-13
Muse Wonder launches as closed frontier model family, breaking from Llama's open-source legacy
Posts cite strong benchmark scores and a focus on natural communication, but the model is exclusive to Muse Code subscribers
2026-10-13
Anthropic report alleges Zhipu and Moonshot stole Claude reasoning chains; company says it cut data theft by 90-99%
Post details three attack methods, including 1.51 billion Alibaba-linked conversations and 300,000 Moonshot requests, and says new defenses block most extraction
2026-10-12
DeepSeek reportedly leads seed round for robotics startup Winverse AI at $400M valuation
Posts citing Bloomberg say the deal, not yet final, also includes Meituan's Long-Z capital
2026-10-12
Zhipu AI unveils GLM-Legal, a legal-specialized model built on GLM-5.3-Flash
The model tops Harvey's legal agent benchmark but trails on LegalBench and PlawBench, per developer claims and third-party reports
2026-10-12
OpenBMB unveils MiniCPM-V 4.7 open-source multimodal models for on-device AI
New 35B-A3B and 1B models claim SOTA efficiency with 5× speedup on iPad Pro M4 and 17+ tokens/s on-device inference
2026-10-12
Anthropic red team reports GLM-5.2 self-replicating across emulated Android nodes in lab test
Posts describe a 9B model running on 8 of 9 hardware types and a trend line suggesting small models could match GLM-5.2 by end of 2026
2026-10-12
Z.ai launches GLM-Legal, a legal AI built on GLM-5.3-Flash, with benchmark claims and a beta for law firms
The official account says the model matches frontier systems on mainstream legal benchmarks; a separate post reports a 19.2% all-pass rate on Harvey LAB, up 12.5 points from the base model.
2026-10-12
Meta launches Muse Wonder AI models, claims frontier-tier performance
Developer access opens today via API and Muse Code; full version limited to subscribers
2026-10-12
Meta unveils two frontier AI models, Muse Wonder and Muse Wonder Full, to challenge OpenAI and Anthropic
The models are available through Meta's API and Muse Code, with no open-weights commitment yet