⏰ Limited-Time Free Models
Several limited-time free models are expiring soon on OpenRouter:
| Model | Expires | Pricing (free tier) | Context |
|---|---|---|---|
| Nex AGI: DeepSeek V3.1 Nex N1 | Jun 8 (1 day) | Free until expiry | 128K |
| NVIDIA: Nemotron Nano 9B V2 | Jun 11 (4 days) | Free until expiry | 128K |
| DeepSeek: R1 Distill Llama 70B | Jun 11 (4 days) | Free until expiry | 128K |
| Z.ai: GLM 4.5 | Jun 19 (12 days) | Free until expiry | 128K |
| Anthropic: Claude Opus 4.6 (Fast) | Jun 29 (22 days) | Free until expiry | 1M |
Set your API calls to fallback gracefully when these expire or migrate to the paid tiers listed below.
🚀 Anthropic Files for IPO, Calls for Global AI Pause
Anthropic confidentially filed for its highly anticipated IPO this week, following a $65 billion funding round at a $965 billion valuation. The company’s S-1 filing marks one of the largest tech IPOs in history alongside SpaceX and OpenAI.
In a striking simultaneous announcement, Anthropic published a white paper calling for a global pause on frontier AI development, warning that “recursive self-improvement” — where AI models improve themselves without human intervention — could arrive sooner than expected and create existential risks. The company urged governments to establish a framework for pausing development of models above certain capability thresholds.
Anthropic is also facing a Pentagon standoff over its refusal to sign certain military AI contracts, with the U.S. government in court over the dispute. Separately, Anthropic is paying xAI $1.25 billion per month for compute infrastructure — a deal reportedly disclosed in SpaceX’s IPO filing.
🏛️ Trump Signs Scaled-Back AI Executive Order
President Trump signed an executive order establishing a voluntary federal framework for vetting advanced AI models for national security risks. The order requires AI companies to submit frontier models for government review before public release — but on a voluntary, not mandatory, basis.
The scaled-back order represents a compromise after earlier drafts proposed more aggressive regulation. OpenAI confirmed it will comply with the order. The White House also ordered federal agencies to stop using Anthropic products, deepening the administration’s rift with the AI company.
🇨🇳 DeepSeek Unveils V4 Preview, Raises $7.4B
Chinese AI lab DeepSeek released a preview of its DeepSeek V4 model, claiming it “closes the gap” with frontier Western models. Early benchmarks show competitive performance on coding and reasoning tasks.
DeepSeek also made its 75% price cut on V4-Pro permanent — now priced at $0.50/M tokens input and $2/M tokens output — dramatically undercutting Western rivals. In a major funding move, the startup is raising $7.4 billion at a $59 billion valuation, with reports suggesting Tencent is among the lead investors. The Economist notes V4 “failed to impress” some analysts, but the aggressive pricing strategy signals China’s intent to lead on cost-efficient AI.
🔧 Microsoft Build 2026: Own AI Models, Scout Personal Agent
At Microsoft Build 2026, the company unveiled its own family of advanced reasoning AI models, signaling reduced reliance on OpenAI. The new MAI (Microsoft AI) models include a frontier reasoning model that competes with GPT-5.5 and Claude Opus 4.8.
Key announcements:
- Microsoft Scout: A persistent “always-on” personal agent integrated into Windows
- Microsoft Fabric: Agentic apps and databases for enterprise AI
- Security: Code, agent, and model security across the development lifecycle
- Healthcare: A frontier AI model co-developed with Mayo Clinic
LinkedIn co-founder Reid Hoffman also departed Microsoft’s board after nearly a decade.
💻 Nvidia GTC Taipei: Record $58.3B Profit, New PC Chips
At GTC Taipei 2025 during COMPUTEX, Nvidia CEO Jensen Huang announced new PC chips that represent the company’s bid to dominate every layer of the AI stack — from data center GPUs to consumer devices.
Nvidia posted a record $58.3 billion profit in its latest quarter, with revenue tripling year-over-year. Huang predicted further growth and warned that “the cost of compute is far beyond the costs of employees” — a statement that went viral as some companies find AI more expensive than human labor.
The company has already committed $40 billion to equity AI deals this year. Highlights include the launch of Nemotron 3 Ultra (America’s best open AI model per some benchmarks), and a partnership with Microsoft to reinvent Windows PCs for the AI era.
🤖 Meta Pivots to Proprietary Muse Spark, Delays Llama Successor
Meta has abandoned its open-source Llama strategy, shifting to a proprietary model called Muse Spark — its first since forming the Superintelligence Labs division with Alexandr Wang. The company reportedly delayed Llama’s successor after performance concerns, with the New York Times reporting internal benchmarks fell short of expectations.
Meta simultaneously slashed 8,000 jobs in an AI-focused reorganization and faces a lawsuit from publishers for allegedly training Llama on pirated books. Meanwhile, a tiny startup called Arcee AI built a 400B-parameter open-source LLM from scratch that reportedly beats Meta’s Llama on key benchmarks.
🇪🇺 SoftBank Commits €75B to French AI Data Centers
SoftBank CEO Masayoshi Son announced plans to invest up to €75 billion ($87.3 billion) to build 5 GW of AI data center capacity in France, as part of the Choose France summit that secured a record €93 billion in total AI-related investment. Son, now Asia’s richest person, said the AI revolution is “50× bigger” than the dot-com boom.
📋 In Brief
- Moonshot AI released Kimi K2.6 with 1 trillion parameters and attention optimizations, while closing a $2 billion funding round at a $20 billion valuation. The company is reportedly considering a Hong Kong IPO.
- xAI launched Grok Build 0.1, its first AI coding agent, alongside Grok Imagine 1.5 Preview for image generation. A new model reportedly tripling parameters is expected by mid-June.
- Alibaba opened its Qwen AI platform to third-party services (KFC, Luckin Coffee, airlines) and released Qwen3.7-Plus with vision, deep reasoning, and autonomous iteration capabilities. The company also revealed a more powerful Zhenwu AI chip.
- Mistral AI expanded into industrial applications, partnering with Airbus and BMW, acquiring an Austrian physics AI startup, and exploring custom chip designs. The company also launched its Vibe platform and defended its military AI work.
- OpenAI brought GPT-5.5 and Codex to Amazon Bedrock, upgraded ChatGPT’s memory architecture, and faces a Florida lawsuit linking ChatGPT to a school shooting case.
- AI CEOs (OpenAI, Anthropic, Microsoft) jointly warned Congress that AI makes it too easy to design bioweapons, urging stricter oversight.
- US House lawmakers released a bipartisan draft bill that would preempt state AI regulations for three years.
- NSA published security guidance for designing AI systems using the Model Context Protocol (MCP), while The Hacker News reported an Anthropic MCP design vulnerability enabling RCE.
- Pentagon is testing rival AI models to potentially replace Anthropic, with its AI user base hitting 1.5 million.
- AirTrunk committed $30 billion to build 5 GW of AI data centers in India.
🤗 Trending on Hugging Face
Hugging Face activity was relatively quiet today, with mostly experimental or low-download uploads. Notable items:
- DevQuasar/nvidia.NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16-GGUF (9 downloads): A GGUF quantized version of NVIDIA’s Nemotron 3 Ultra, compatible with local inference frameworks.
- mradermacher/Gemma-4-12B-it-AEON-Abliterated-K4-BF16-GGUF: An abliterated GGUF variant of Google’s Gemma 4 for local deployment.
Most other uploads showed zero downloads and no meaningful tags — typical ecosystem noise.
🚀 New Models on OpenRouter
- NVIDIA Nemotron 3.5 Content Safety (free): 128K context, free tier available — a specialized safety classifier model.
(NVIDIA Nemotron 3 Ultra and Qwen3.7-Plus are covered in the news sections above.)
⭐ GitHub Trending: AI Edition
| Repo | Stars | Description |
|---|---|---|
| pewdiepie-archdaemon/odysseus | 54,476 ★ | Self-hosted AI workspace |
| asz798838958/aBaiAutoplus | 1,524 ★ | Multi-platform AI account auto-registration & ChatGPT Plus management |
| ClaudioDrews/memory-os | 856 ★ | 7-layer memory OS for Hermes Agent with Qdrant persistent memory |
| jd-opensource/JoyAI-Echo | 530 ★ | Pushing the frontier of long audio-visual generation |
| liyue-aigc/female-portrait-director | 517 ★ | Modular AI portrait prompt engineering toolkit |