⏰ Limited-Time Free Models
Four free-tier models are nearing expiration on OpenRouter. Grab them while they last:
| Model | Expires | Context | Prompt/Completion (per M tokens) |
|---|---|---|---|
| Baidu: Qianfan-OCR-Fast | May 28 (1 day) | 65K | $0.68 / $2.81 |
| Google: Gemini 2.0 Flash Lite | Jun 1 (5 days) | 1M | $0.075 / $0.30 |
| Google: Gemini 2.0 Flash | Jun 1 (5 days) | 1M | $0.10 / $0.40 |
| Mistral: Mistral 7B Instruct v0.1 | May 30 (3 days) | 4K | $0.11 / $0.19 |
🚀 DeepSeek Makes 75% V4-Pro Price Cut Permanent
DeepSeek has permanently slashed prices on its flagship V4-Pro model by 75%, escalating the global AI pricing war. The move makes V4-Pro the best bang-for-buck model on the market — output tokens are now priced at least 34× below GPT-5.5, according to comparisons by the-decoder.com. Reuters reports the price cut comes as DeepSeek eyes a new funding round, while the South China Morning Post notes the model now tops global value rankings. The steep discount locks in a price advantage that challenges US AI leaders on cost.
📜 Pope Leo XIV Issues Landmark AI Encyclical
Pope Leo XIV released his first encyclical — a sweeping AI manifesto titled Magnifica Humanitas — warning that some AI-driven weapons are now “practically beyond human control.” The document calls for robust international regulation of artificial intelligence, denounces “opaque algorithms” that concentrate power, and urges a human-centered approach to technological development. The encyclical draws comparisons to past papal interventions on nuclear weapons and climate change, with the Pope specifically calling to “disarm” AI and warning of a new “Tower of Babel.” The Vatican’s intervention adds a powerful moral voice to the global AI governance debate.
🏛️ White House Approves $9 Billion for Spy Agency AI
The White House has approved a $9 billion plan to modernize US intelligence agencies with cutting-edge AI capabilities, according to the New York Times. The secret initiative is driven by concerns that American spy agencies are falling behind in AI adoption compared to adversaries. The funds will accelerate deployment of Nvidia AI chips across intelligence infrastructure and fund next-generation analytics platforms. The move signals that AI is now viewed as a national security imperative at the highest levels of government.
📚 Wikipedia Signs AI Data Deals with Tech Giants
On its 25th anniversary, the Wikimedia Foundation announced paid data access agreements with Microsoft, Meta, Amazon, and Perplexity — marking a major shift in how Wikipedia monetizes its vast corpus. The deals provide AI companies with priority access to Wikipedia’s structured data for training large language models. The agreements come as Wikipedia faces declining organic traffic and seeks sustainable revenue streams. Forbes notes the move parallels similar content licensing deals struck by Reddit and Stack Overflow with AI companies.
🔄 Meta Launches Muse Spark, Abandons Open-Source Llama
Meta has released Muse Spark, its first proprietary AI model under the newly formed Superintelligence Labs division, signaling a decisive break from the open-source Llama series. The model represents Meta’s rebuilt AI stack after Llama’s underwhelming performance. VentureBeat reports Muse Spark is Meta’s first public model since hiring Alexandr Wang as chief AI scientist. The shift to closed-source has drawn criticism — The Register notes the model is “as open as Zuckerberg’s private school” — but Wall Street reacted positively, with Meta’s stock surging on the release. Meanwhile, five major publishers have filed a copyright lawsuit against Meta, alleging Llama was trained on unauthorized copyrighted works.
🔒 Anthropic Investigates Unauthorized Mythos Access
Anthropic is investigating reports that a rogue group gained unauthorized access to its restricted Mythos AI model, which the company had withheld from public release due to cybersecurity risks. Early reports from WIRED and Bloomberg suggest Discord-based sleuths may have found a way through the safeguards. The incident has sparked alarm in Washington — CNBC reports that Federal Reserve Chair Powell and Treasury Secretary Bessent discussed the Mythos threat with major US banks. The breach, if confirmed, would be one of the most serious AI security incidents to date, given Mythos’s demonstrated ability to identify thousands of security vulnerabilities.
🏢 OpenAI Nears Historic Trillion-Dollar IPO
OpenAI is preparing to confidentially file for an IPO that could value the company at up to $1 trillion, according to sources cited by CNBC and Reuters. The filing could come within weeks, making it one of the largest public offerings in history. Fortune notes the IPO will finally reveal OpenAI’s closely guarded financials, while questions remain about the company’s path to sustainable profitability amid massive compute expenditures. The IPO comes alongside SpaceX’s own blockbuster filing, creating what analysts call an unprecedented “mega-IPO” window.
🤗 Trending on Hugging Face
- OpenTransformer/AGILLM-4 — A new long-context language model based on the AGILLM architecture, designed for extended reasoning tasks.
- tjarvis91/vfai-x-3.5-9b-options — A Qwen 3.5-based image-text-to-text model fine-tuned for options trading analysis (1,085 downloads).
- rainbowrobotics/pi05_0412_abs — A robotics model in the pi0.5 series, trained on the simtos_0412_abs dataset using LeRobot.
- djaldjoul/silma9b-algerian-medical — A 9B-parameter SFT model for Algerian medical domain applications, built with Unsloth.
⭐ GitHub Trending: AI Edition
| Repository | Stars | Language | Description |
|---|---|---|---|
| VILA-Lab/FigMirror | 279 | Python | AI agent that plots your data in any academic paper’s figure style |
| XingYu-Zhong/DeepSeek-GUI | 259 | TypeScript | AI agent workspace for DeepSeek models with Code and Claw modes |
| akitaonrails/ai-memory | 221 | Rust | Long-term memory solution for agent coding CLIs, enabling handoff between different agent vendors |
| edgenuitybot/exodusbot | 271 | JavaScript | Automated quiz/testing bot (education sector) |
| CylinderChairman/aida64 | 221 | PowerShell | AIDA64 hardware diagnostic tool |
💡 Key Trends
- Pricing war enters new phase: DeepSeek’s permanent 75% cut on V4-Pro signals that Chinese AI labs are competing aggressively on cost, forcing US providers to respond or lose market share.
- AI safety and governance go mainstream: From the Pope’s encyclical to White House intelligence funding to the Mythos breach investigation, AI governance is no longer a niche concern — it’s a top-tier global policy issue.
- The open-source pivot: Meta’s shift from Llama to proprietary Muse Spark marks a major recalibration. The industry is splitting into two camps: companies that see open-weight models as strategic assets and those retreating to closed-source for competitive advantage.
- Data becomes a revenue stream: Wikipedia’s AI data licensing deals with five tech giants follow the pattern set by Reddit and Stack Overflow — user-generated content is becoming a valuable training data commodity.
- Agent memory and tooling: The rise of projects like ai-memory (Rust-based agent handoff) and FigMirror (agent-driven visualization) shows the ecosystem is maturing beyond chat into specialized agent capabilities.