⏰ Limited-Time Free Models Expiring Soon

Several premium models remain available for free on OpenRouter with limited-time offers. Notable expirations this week:

Model Context Pricing Expires
Nex AGI: DeepSeek V3.1 Nex N1 131K $0.14/M tokens input, $0.50/M tokens output Jun 8 (tomorrow!)
NVIDIA: Nemotron Nano 9B V2 131K $0.04/M tokens input, $0.16/M tokens output Jun 11
DeepSeek: R1 Distill Llama 70B 131K $0.70/M tokens input, $0.80/M tokens output Jun 11
Z.ai: GLM 4.5 131K $0.60/M tokens input, $2.20/M tokens output Jun 19
Meta: Llama 3 70B Instruct 8K $0.51/M tokens input, $0.74/M tokens output Jun 19
Anthropic: Claude Opus 4.6 (Fast) 1M $30/M tokens input, $150/M tokens output Jun 29

🚨 Anthropic Warns AI May Begin Recursive Self-Improvement, Urges Global Pause

Anthropic published a stark warning that AI systems could soon achieve recursive self-improvement — the ability to build better versions of themselves without human involvement. In a series of posts covered by The New York Times, Reuters, Forbes, CNN, and Scientific American, the company urged AI labs worldwide to pause development on advanced models until safety frameworks are in place. An Anthropic blog post called for “AI nonproliferation” as a global priority, arguing that the next stage of AI evolution — machines that design and build their own successors — poses existential risks if left unchecked.

🏛️ Trump Signs Executive Order for Voluntary Federal AI Model Vetting

President Trump signed an executive order that establishes a voluntary framework for federal oversight of frontier AI models. Reported by PBS, The New York Times, and Politico, the order allows top AI developers to submit their models for national security risk assessment by the federal government. While voluntary, the order creates incentives for participation and signals a shift toward structured government engagement with frontier AI — a notable contrast to earlier administration tensions with companies like Anthropic.

🔒 OpenAI Unveils Lockdown Mode; White House Considers Equity Stake

OpenAI released Lockdown Mode, a new security feature designed to protect sensitive data from prompt injection attacks, as reported by TechCrunch. The feature allows enterprises to deploy ChatGPT and API models in high-security environments where data leakage risks are a top concern. Meanwhile, TechCrunch also reported that the Trump administration is evaluating taking an equity stake in OpenAI as part of a broader partnership discussion, marking an unprecedented level of government involvement in a private AI company. Separately, Axios revealed that Sam Altman shared insights on OpenAI’s top token user.

💰 Nvidia Posts Record $58.3B Profit, $40B in AI Equity Deals

Nvidia reported a record $58.3 billion quarterly profit amid the ongoing AI chip boom, Al Jazeera reports. At GTC Taipei 2026, CEO Jensen Huang delivered a keynote covering the company’s roadmap and vision for AI infrastructure, while TechCrunch revealed that Nvidia has already committed $40 billion to equity AI deals this year. A Fortune report quotes an Nvidia executive noting that “the cost of compute is far beyond the costs of the employees” — highlighting the immense capital required for AI inference at scale.

🌐 Alibaba Opens Qwen to Third-Party Agents; New Models Top Arena Rankings

Alibaba has opened its Qwen AI platform to third-party services, including KFC, Luckin Coffee, and airlines, in a major push for AI agent dominance (Caixin, CNBC). The expansion brings Qwen-powered agentic shopping to Taobao and allows voice-controlled car interactions. Separately, Alibaba teased new Qwen preview models that have become the highest-ranking Chinese AI models on the Chatbot Arena leaderboard (SCMP). The Qwen team also entered embodied AI with the Qwen-VLA model for vision-language-action reasoning (Pandaily).

💻 Moonshot AI Releases Kimi-K2.6 (1T Parameters), Code CLI, Raises $2B

Moonshot AI released Kimi-K2.6, a 1-trillion-parameter open-source model with attention optimizations that has driven significant traffic on OpenRouter (SiliconANGLE, Tech Times). The company also launched Kimi Code CLI, a terminal-based AI coding agent built in TypeScript (MarkTechPost). On the business front, Moonshot raised $2 billion at a $20 billion valuation (TechCrunch) and is reportedly considering a Hong Kong IPO as AI stocks flourish (Bloomberg, SCMP).

In Brief

  • DeepSeek made a permanent 75% price cut on its flagship V4-Pro model (Reuters, InfoWorld), and is pursuing a $7.4 billion maiden funding round at a ~$59 billion valuation (Indiatimes).
  • Microsoft Build 2026 delivered 7 major AI announcements, including new in-house AI models to lessen reliance on OpenAI and lower costs for developers (CNBC, Microsoft Blog).
  • Google launched the Gemini Enterprise Agent Platform, expanding Vertex AI into a full agent stack (HPCwire). Separately, Google will pay SpaceX $920 million per month for compute capacity at xAI data centers (TechCrunch, CNBC).
  • Meta slashed 8,000 jobs as it pivots toward AI (NPR) and launched Muse Spark, a proprietary AI model replacing its open-source Llama strategy (VentureBeat, Forbes).
  • Mistral AI acquired an Austrian physics AI startup in an industrial push (Reuters) and is expanding into custom chip designs.
  • ByteDance is weighing up to $70 billion in 2026 AI capex (Bloomberg); Qualcomm signed a major AI chip deal with ByteDance (Reuters, 24/7 Wall St.).
  • SoftBank committed €75 billion ($87.3B) to French AI data centers (Euronews, Reuters); Masayoshi Son became Asia’s richest person amid the AI boom (Forbes).
  • The US House unveiled a bipartisan draft bill to preempt state AI laws for three years (Reuters, Politico, Axios).
  • A new bipartisan bill aims to regulate military uses of AI (Defense One, NYT).
  • Sriram Krishnan departed his role as White House AI advisor (TechCrunch).

🤗 HuggingFace Highlights

The HF ecosystem saw a relatively quiet day today with several experimental uploads:

⭐ GitHub Trending: AI Edition

  • jd-opensource/JoyAI-Echo — Pushing the frontier of long audio-visual generation. JD.com’s open-source team releases a system for extended video with synchronized audio (811 ★, Python)
  • tastyeffectco/sandboxes — Self-hosted dev sandboxes with instant preview URLs, designed for coding agents. One command, no Kubernetes (485 ★, Go)
  • CWS6206/ai-coding-starter-kit — Curated agent skills, checklists, and templates for Swiss development teams using AI-assisted coding workflows (261 ★)
  • razr001/align-dev — Helps AI-assisted frontend teams generate shared coding standards and SKILL.md files so agents produce consistent code (258 ★, TypeScript)

💡 Key Trends

This session’s news paints a Sunday of extraordinary density. Four themes stand out. First, the Anthropic recursive self-improvement warning — covered across NYT, Reuters, Forbes, CNN, and Scientific American — signals that AI safety has moved from abstract debate to urgent, multi-stakeholder concern with concrete calls for global pauses and nonproliferation treaties. Second, government engagement with AI is accelerating on multiple fronts: Trump’s voluntary EO for model vetting, the White House considering an equity stake in OpenAI, a House draft bill preempting state AI laws, and a new bipartisan military AI regulation bill. Third, the massive capital flows — Nvidia’s $58.3B profit and $40B in equity deals, DeepSeek’s $7.4B raise, SoftBank’s €75B French investment, ByteDance’s $70B capex — underscore that AI infrastructure spending is now measured in national-budget terms. Fourth, the platform battle for AI agents is intensifying: Alibaba opening Qwen to third-party services, Google’s Enterprise Agent Platform, and Moonshot’s Kimi-K2.6 leading OpenRouter traffic, all competing for the same enterprise and consumer touchpoints.