🏛️ OpenAI to File for Trillion-Dollar IPO as Soon as Friday
OpenAI is preparing to confidentially file for an initial public offering as soon as this Friday, according to CNBC and Reuters sources. The IPO could value the company at over $1 trillion, making it one of the largest public market debuts in history — alongside SpaceX, which also unveiled its IPO filing this week. Fortune examines the big questions the filing may finally answer about OpenAI’s finances, governance, and path to AGI. In a parallel milestone, an OpenAI model has disproved a central conjecture in discrete geometry — an 80-year-old unsolved math problem — reported by The Guardian and OpenAI’s own blog.
🔓 Anthropic’s Mythos AI Reportedly Accessed by Unauthorized Users — $1.25B/month xAI Compute Deal Revealed
A tumultuous day for Anthropic on multiple fronts. Reports from Bloomberg, TechCrunch, and BBC indicate that unauthorized users have gained access to Anthropic’s powerful Mythos AI model, a cybersecurity tool so potent that Anthropic itself had withheld it from wider release. The breach — which Anthropic is investigating — has prompted discussions between Treasury Secretary Powell, National Economic Advisor Bessent, and major U.S. banks (CNBC). Meanwhile, SpaceX’s IPO filing has revealed that Anthropic is paying $1.25 billion per month ($15B/year) to xAI for data center compute capacity (WIRED, Axios, TechCrunch) — one of the largest compute deals in the industry’s history.
🔄 Meta Abandons Open-Source Llama, Launches Proprietary Muse Spark
Meta has officially launched Muse Spark, its first proprietary AI model since the formation of its Superintelligence Labs under Alexandr Wang (Fortune, VentureBeat, Reuters). The move marks a definitive break from Meta’s open-source Llama lineage — reports from The New Stack, Forbes, and CNBC indicate Meta has delayed or abandoned Llama successors in favor of the closed-source Muse Spark architecture. Muse Spark emphasizes voice interactions and shopping integration for Meta’s Indian market launch (Deccan Herald). The strategic pivot comes as Meta faces multiple copyright lawsuits over Llama’s training data and growing investor pressure to monetize its AI investments.
🌍 Google Debuts Gemini Omni World Model at I/O 2026
At Google I/O 2026, Google unveiled Gemini Omni — a new world model with advanced AI video capabilities (Mashable, PCMag, CNET). The announcement was part of a sweeping set of I/O releases including a fundamental redesign of Google Search powered by AI (NYT: “Google Changes Its Search Box for the First Time in 25 Years”). Google also introduced updates to the Gemini app positioning it against ChatGPT and Claude, smart home integration with Gemini, and over 100 other announcements. The Omni model signals Google’s ambition to lead in multimodal and world-modeling AI.
💰 Moonshot AI Raises $2B at $20B Valuation, Releases Kimi K2.6 with 300-Agent Swarms
China’s Moonshot AI has raised $2 billion in a funding round led by Meituan, valuing the Kimi chatbot maker at over $20 billion (Bloomberg, TechCrunch, Forbes). On the same day, the company released Kimi K2.6 — a 1-trillion-parameter model featuring long-horizon coding, agent swarm scaling to 300 sub-agents, and 4,000 coordinated steps (SiliconANGLE, MarkTechPost). CoreWeave and Cerebras reported inference speeds of 981 tokens/second on K2.6, 6.7× faster than GPU cloud alternatives. The dual announcement positions Moonshot AI as a major challenger to both U.S. frontier labs and Chinese rivals like Alibaba and DeepSeek.
📈 Micron Joins $1 Trillion Club; S&P 500 and Nasdaq Hit Record Highs on AI Optimism
Micron Technology has crossed the $1 trillion market capitalization mark, joining the elite club of AI-driven tech giants (Forbes, TradingView). The leap is fueled by surging demand for HBM memory chips powering AI data centers. The broader market reflected AI optimism — the S&P 500 and Nasdaq hit record closing highs, with Goldman Sachs estimating AI capex will reach $800 billion in 2026 (Benzinga, Yahoo Finance). Bank of America sees AI infrastructure spending topping $1 trillion by 2027. The numbers underscore an investment cycle without modern precedent.
💸 DeepSeek Makes 75% Price Cut Permanent; Pricing War Intensifies
DeepSeek has locked in its 75% price reduction on the flagship V4-Pro model as a permanent change — escalating the frontier AI pricing war (Reuters, Engadget, Bloomberg). Output tokens are now priced at levels that NIST evaluations rank as the top bang-for-buck globally, putting sustained pressure on OpenAI and Anthropic’s pricing models.
⏰ Limited-Time Free Models Expiring Soon
| Model | Expires | Pricing |
|---|---|---|
| Baidu: Qianfan-OCR-Fast | 2026-05-28 | $0.68/M tokens prompt · $2.81/M tokens completion |
| Mistral: Mistral 7B Instruct v0.1 | 2026-05-30 | $0.11/M tokens prompt · $0.19/M tokens completion |
| Google: Gemini 2.0 Flash Lite | 2026-06-01 | $0.075/M tokens prompt · $0.30/M tokens completion |
| Google: Gemini 2.0 Flash | 2026-06-01 | $0.10/M tokens prompt · $0.40/M tokens completion |
| Qwen: Qwen3 30B A3B | 2026-06-05 | $0.09/M tokens prompt · $0.45/M tokens completion |
| Sao10k: Llama 3 Euryale 70B v2.1 | 2026-06-05 | $1.48/M tokens |
| NousResearch: Hermes 2 Pro - Llama-3 8B | 2026-06-05 | $0.14/M tokens |
🤗 Trending on Hugging Face
- L1nus/qwen3-4b-pubmedqa-thinking-default — Qwen3 4B fine-tuned on PubMedQA with chain-of-thought reasoning using Unsloth and TRL.
- OpenTransformer/AGILLM-4 — Long-context language model in the AGILLM series (PyTorch, transformer architecture).
- redue/qwen-chan3.5-gguf — Qwen 3.5 GGUF-quantized model for llama.cpp with vision-language capabilities.
- smdesai/supertonic-3-coreml — Text-to-speech model for speech synthesis (ONNX, CoreML).
- al-12/qwen-image-edit-lightning-endpoint — Qwen-based image editing endpoint deployment.
⭐ GitHub Trending: AI Edition
- VILA-Lab/FigMirror ⭐308 — An automated AI agent tool for plotting data in any paper’s figure style (Python).
- XingYu-Zhong/DeepSeek-GUI ⭐307 — AI agent workspace for DeepSeek models with Code and Claw modes (TypeScript).
- study8677/awesome-architecture ⭐287 — 21 architecture maps covering AI gateway, RAG, and microservices (Vue).
- akitaonrails/ai-memory ⭐281 — Long-term memory solution for agent coding CLIs and agent handoff, built in Rust.
💡 Key Trends
This edition captures a market at an inflection point: public markets embracing AI at unprecedented scale (OpenAI’s trillion-dollar IPO, Micron joining the $1T club, Nasdaq records); security and governance catching up to capability (Mythos access breach, NSA MCP guidance, Meta’s copyright lawsuits); and the compute arms race intensifying (Anthropic’s $15B/year xAI deal, $800B+ global capex, Moonshot AI’s $2B raise). The juxtaposition of Anthropic restraining Mythos due to safety concerns while simultaneously spending $15B/year on compute captures the central tension of the industry in 2026.