ChatGPT for Teens & Anthropic $65B: AI News Aug 19, 2026

1. Top Headlines

• OpenAI launches ChatGPT for Teens — a global rollout (starting Aug 18) for users aged 13–17 with stronger guardrails on self-harm, violence and eating-disorder topics, plus Study Mode (guided problem-solving, quizzes, visualizations), Homework reminders, and Quiet Hours that parents or teens can set. OpenAI will use self-reported age plus behavioral and account signals to route minors into the experience. (Source: OpenAI news, Aug 18)
• Anthropic says revenue run rate hit $65B — in an investor update (weekend of Aug 15–17), Anthropic told investors its annualized revenue run rate reached about $65 billion by end of July 2026, up from ~$9B at end-2025 and ~$47B in May; it also reported preliminary Q2 2026 revenue above $11.5B with positive adjusted operating income — its first profitable quarter on that measure. (Source: Bloomberg, CNBC, Reuters, Axios, Aug 17; subject to vendor confirmation)
• Anthropic rolls out invisible Claude watermark globally — following its Aug 14 explainer, all Claude models launched from Aug 2, 2026 embed an imperceptible SynthID-style text watermark and C2PA provenance metadata for images, applied globally (not just the EU) to meet the EU AI Act's Article 50 transparency rule. (Source: Anthropic news, Aug 14)

2. Model Release Dynamics

• xAI / SpaceXAI Grok 4.6 (Aug 12) — Artificial Analysis Intelligence Index 61, priced at $2 / $6 per 1M tokens with 500K context; trails Claude Fable 5 by one point but beats GPT-5.6 Sol on two coding benchmarks. Musk says Grok 4.7 could arrive within 3–4 weeks. (Source: xAI, Artificial Analysis; subject to vendor confirmation)
• DeepSeek V4 Pro stable build (0813) — output at $0.87 / 1M tokens (≈1/7 of Grok 4.6), 1M context, Terminal-Bench 2.1 87.9 (0.1 behind Fable 5), DeepSWE 62.7 (up from 12.8); open weights plus the open-source Harness agent framework. Peak-valley pricing took effect Aug 17. (Source: DeepSeek, Aug 13; subject to vendor confirmation)
• Qwen3.8 Max open weights (Aug 12–13) — Alibaba published full 2.4T-parameter (A95B active) Max-tier weights on Hugging Face and ModelScope, its first open Max flagship; leads PaperBench (93.0) and OSWorld (86.1), trails on SWE-bench Pro (67.7). (Source: Alibaba Cloud, Aug 13; subject to vendor confirmation)
• Zhipu GLM-5.3 (Aug 14) — 753B open-weight flagship the company calls its strongest open-source coding model; Terminal-Bench 3.0 28.3, DeepSWE v1.1 66.9, CyberGym 84.5. (Source: Zhipu technical report, Aug 14; subject to vendor confirmation)
• NVIDIA Nemotron 3.5 Lightning (Aug 11) — open 30B MoE (≈3B active) tuned for high-volume agent execution, up to 4x output speed vs similar models, 1M context, runs on a single H100/DGX Spark; ships with the NeMo Switchyard routing library. (Source: NVIDIA, Hugging Face, Aug 11)

3. Industry & Capital

• Anthropic IPO timeline — having confidentially filed an S-1 in June, Anthropic is preparing a likely fall listing; investors cite a possible ~$2T valuation (above SpaceX's $1.77T record), though that rests on unaudited figures and no net profit yet. (Source: Bloomberg, FT, Aug 17; subject to vendor confirmation)
• OpenAI's enterprise surge — CFO Sarah Friar told investors enterprise revenue has overtaken consumer ChatGPT revenue (a 60/40 split at the start of 2026 has crossed), with ~$40B annualized revenue, July +20% MoM, enterprise customers +32%; ads approaching $1B annualized. (Source: CNBC, Aug 15; subject to vendor confirmation)
• NVIDIA's $500B compute-financing platform — non-binding MOUs with Apollo, Blackstone, Brookfield, BlackRock, Goldman Sachs and KKR to fund customer AI-compute builds, alongside the open Nemotron 3.5 Lightning push into agent cost-efficiency. (Source: Les Échos, Aug 11; subject to vendor confirmation)
• Databricks ~$190B, Cognition ~$40B — AI-data and agent valuations keep climbing as the IPO window opens. (Source: industry reports, Aug 13–14; subject to vendor confirmation)

4. China's AI Momentum

• Open-source lead widens — Hugging Face's spring report puts China's open-model download share at 41%, first ahead of the US; Chinese open models hold OpenRouter's top-4 weekly usage slots. The Aug 13 cluster — DeepSeek V4 Pro, Qwen3.8 Max weights, GLM-5.3 — compressed a quarterly cadence into one day. (Source: cinic.org.cn, Aug 14; subject to vendor confirmation)
• DeepSeek V4 Pro peak-valley pricing (Aug 17) — off-peak roughly half of peak, easing load and pulling heavy jobs into idle hours; output cost is a fraction of US frontier models. (Source: DeepSeek, Aug 14; subject to vendor confirmation)
• Tencent WorkBuddy #1 domestically — 20.97M monthly PC visits in Q2, ahead of ByteDance's TRAE and Alibaba's QoderWork; Q2 capex RMB 52.8B ($7.3B), +64% vs consensus, pushing FCF negative as it builds Hunyuan and inference compute. (Source: ChinaBiz Insider / JPMorgan, Aug 13; subject to vendor confirmation)
• Stanford AI Index 2026 — US–China model-performance gap now "essentially disappeared," the two alternating at the lead since early 2025. (Source: Stanford HAI, 2026; subject to vendor confirmation)

5. Editor's Take

OpenAI's teen mode and Anthropic's $65B run rate are two sides of the same coin: the consumer chatbot is being turned into a regulated, educational utility, while the real money has already moved to enterprise API and coding agents. I'm more struck by the watermark story — Anthropic is exporting an EU regulation into global product behavior, and OpenAI's refusal to watermark ChatGPT text (citing false-positive risk) sets up the first real fork in AI provenance. The open-weight wave from China isn't just cheaper; it's redefining what "open source" means, and the West's pricing floor may not survive the year.