Google launches Gemini 4 Argon

📰 Today's Headlines

Google releases Gemini 4 Argon at $2 per million input tokens (5 min read) Google DeepMind's Gemini 4 Argon is now globally available through the API in AI Studio and the Gemini platform, and to Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search. Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95% off. Google says Argon ranks #1 on the Vals Index and that internal teams using it in Antigravity increased agentic code submissions by more than 35% while cutting rollbacks by nearly a third. Read more

DeepSeek reportedly nearing ~100B yuan round at 500B yuan pre-money valuation (4 min read) According to Caijing, DeepSeek is about to complete a round of nearly 100 billion yuan at a pre-money valuation of about 500 billion yuan. Tencent and CATL are leading the round, and new investors include Stony Creek Capital, Legend Capital and Geely. DeepSeek originally planned to raise 50 billion yuan. CATL plans to invest about 10 billion yuan more, on top of the roughly 5 billion yuan it put in during the first round. Why it matters for Asia: Caijing reported in August that the Chinese AI lab is preparing to file for a Sci-Tech Innovation Board listing by year-end and aims to list in 2027. Read more

Anthropic warns small AI models could self-replicate online by end-2026 (3 min read) In a post late Monday, Anthropic said its experiments showed that models with 30 billion parameters or fewer can already perform many of the individual tasks needed for self-replication, though they struggle to complete the full process reliably. It said the models replicated across a simplified test network and could reach autonomous replication by the end of 2026 if smaller models keep improving at their historical rate. Ciaran Martin, a former head of Britain's National Cyber Security Centre, called the post an "interesting technical paper with a more than hint of vendor exaggeration." Read more

🔥 Hot on Social

On Hacker News, a team showed how it turned Google's EmbeddingGemma 2 embedding model into an on-device speech-to-text tool. It transcribes an hour of audio in about 13 seconds on a base MacBook Air, with a word error rate of about 5%, compared with 2–3% for leading speech models. The post drew 433 points and 99 comments. link

⚡ Quick Link

Anthropic says new defenses against Claude distillation cut copycat data harvesting by roughly 90–99%, and names Zhipu and Moonshot (Kimi). link