⏰ Limited-Time Free Models
Three free models on OpenRouter are approaching their expiration dates:
| Model | Expires | Days Left | Pricing |
|---|---|---|---|
| Mistral 7B Instruct v0.1 | 2026-05-30 | 12 days | $0.11/M tokens input · $0.19/M tokens output |
| Gemini 2.0 Flash Lite | 2026-06-01 | 14 days | $0.075/M tokens input · $0.3/M tokens output |
| Gemini 2.0 Flash | 2026-06-01 | 14 days | $0.1/M tokens input · $0.4/M tokens output |
🤗 Trending on Hugging Face
Recent uploads on Hugging Face show diverse activity from fine-tuned LLMs to domain-specific models:
- PS4Research/qa-sft-magistral-24b — A Mistral-based 24B parameter model fine-tuned for QA tasks, built with Transformers and Safetensors.
- JMingo/gemma-4-E2B-it-Japanese — A Japanese instruction-tuned variant of Google’s Gemma 4 E2B model, licensed under Apache 2.0.
- mkychsu/semiconductor_graph_preflexor_grpo_2 — A GRPO-trained model for semiconductor graph tasks, showcasing RL-based fine-tuning with TRL.
- ElioChampaney/NOVA-50M — A compact 50M parameter model, receiving its first community likes.
- turtle170/NetTinyANN — An Apache 2.0-licensed tiny ANN model for lightweight applications.
- BlueTriangles/SDXL_Melvina_MacGarlen — An SDXL LoRA for text-to-image generation based on Illustrious XL v2.0.
📰 Headline AI News
DeepSeek V4 Preview Launches at Disruptive Pricing
China’s DeepSeek released a preview of its long-awaited DeepSeek V4 model, claiming near state-of-the-art performance at approximately 97% below OpenAI’s GPT-5.5 pricing. The model runs on Huawei chips with “full support” from the Chinese telecom giant. At $0.14/M tokens for V4 Pro vs. GPT-5.5’s estimated $8.75/M tokens, DeepSeek continues its strategy of aggressive pricing. China’s state AI fund is backing DeepSeek in an up-to-$4 billion funding round, signaling a major push in the US-China AI race. (CNBC · VentureBeat · SCMP)
OpenAI Releases GPT-5.5 as AI Super App Ambitions Grow
OpenAI launched GPT-5.5, described by TechCrunch as bringing the company “one step closer to an AI ‘super app’.” The release comes amid a heated competitive landscape with DeepSeek and Anthropic. OpenAI also purchased a voice cloning AI startup and is reportedly considering legal action against Apple over a strained partnership. The Musk vs. Altman trial has concluded closing arguments and is heading to the jury. (OpenAI Blog · TechCrunch · Fortune)
Anthropic Tests “Mythos” — Claims Most Powerful Model Yet
Anthropic is reportedly testing “Mythos,” which the company claims is its “most powerful AI model ever developed,” with a focus on cybersecurity. Simultaneously, Accenture and Anthropic announced a partnership to help organizations secure and scale AI-driven cybersecurity operations. (Fortune · NYT)
NVIDIA Launches Ising: Open Models for Quantum Computing
NVIDIA released Ising, described as “the world’s first open AI models to accelerate the path to useful quantum computers.” This marks a significant step toward bridging classical AI with quantum computing research. (NVIDIA Newsroom)
Meta Debuts Muse Spark, Its First “People-First” AI Model
Meta launched Muse Spark, its first model “purpose-built to prioritize people.” The Meta AI app climbed to No. 5 on the App Store following the release, signaling strong consumer interest. (Meta Blog · TechCrunch)
Cerebras Surges 89% in IPO Debut
AI chip startup Cerebras rose 89% on its first day of trading, as tech IPOs ramp up amid strong investor appetite for AI hardware companies. The company’s market debut is seen as a bellwether for the AI semiconductor sector. (NYT)
⭐ GitHub Trending: AI Edition
| Repository | Stars | Language | Description |
|---|---|---|---|
| nexu-io/html-anything | ⭐2,425 | HTML | Agentic HTML editor — local AI agent writes HTML, you ship it. 75 skills × 9 surfaces. |
| huangserva/3DCellForge | ⭐2,088 | JavaScript | AI-powered interactive 3D model generation, inspection, and presentation studio. |
| simonlin1212/a-stock-data | ⭐986 | — | China A-Share market data toolkit: 6-layer architecture, 15 endpoints, 7 data sources. AI skill for market data. |
| yaassin12/DeepSeek-V4-Pro-App | ⭐455 | C++ | Desktop app for DeepSeek V4 Pro — 1.6T MoE, 1M token context. |
| sunyouzhi2003/LSFG-Desktop | ⭐454 | C++ | Lossless Scaling 2026: universal frame generation LSFG 3 with X2/X3/X4 multi-frame support. |
💡 Key Trends
Today’s AI landscape is defined by three major themes:
-
Pricing War Heats Up: DeepSeek V4 at 97% below GPT-5.5 pricing is reshaping the economics of frontier AI. The race to offer competitive intelligence at a fraction of the cost is benefiting developers and enterprises alike.
-
Chip Independence and Diversification: DeepSeek’s partnership with Huawei chips and Cerebras’s successful IPO highlight the growing push for AI hardware alternatives beyond NVIDIA’s dominance.
-
Security Becomes Central: Anthropic’s “Mythos” cybersecurity-focused model and the Accenture-Anthropic partnership signal that AI security is transitioning from a research niche to a commercial priority.
-
Local-First AI Tools: The top GitHub repos emphasize local, agentic, and user-controlled AI — from an HTML editor that runs locally to a 3D model generation studio, reflecting a shift toward personal AI tooling.