⏰ Limited-Time Free Models
Several models on OpenRouter are approaching their free-tier expiration:
| Model | Expires | Context | Prompt Price | Completion Price |
|---|---|---|---|---|
| Arcee AI: Trinity Large Preview | May 21 (tomorrow!) | 131K | $0.15/M tokens | $0.45/M tokens |
| Mistral: Mistral 7B Instruct v0.1 | May 30 | 4K | $0.11/M tokens | $0.19/M tokens |
| Google: Gemini 2.0 Flash Lite | June 1 | 1M | $0.075/M tokens | $0.30/M tokens |
| Google: Gemini 2.0 Flash | June 1 | 1M | $0.10/M tokens | $0.40/M tokens |
Arcee AI’s Trinity Large Preview expires tomorrow — grab it while it lasts.
🚀 Anthropic Withholds Mythos Preview Over Cybersecurity Fears
The biggest AI story of the day: Anthropic has decided not to release its Mythos Preview model after internal testing revealed the model was “too powerful” at offensive cybersecurity tasks. According to reports from Axios, the model demonstrated capability gaps that made it exceptionally effective at hacking — raising alarms across the financial sector.
CNBC reported that Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Bessent discussed the Mythos AI cyber threat with major U.S. banks in a closed-door meeting. The New York Times described it as a model that “sets off global alarms,” while The Guardian questioned whether Mythos could be a threat to global cybersecurity.
Anthropic’s decision to withhold its most capable model marks an unprecedented moment in AI safety — a company voluntarily keeping a frontier model from market because of demonstrated risks.
🚀 Google I/O 2026: Gemini 3.5 Flash and the Agentic Era
Google’s I/O 2026 keynote dominated tech headlines with a barrage of AI announcements:
- Gemini 3.5 Flash — Google’s latest frontier model now available on OpenRouter with 1M context at $1.5/M tokens prompt pricing, positioning it as one of the most cost-effective high-capability models on the market.
- “Gemini Spark” — a new personal AI agent capable of proactive, 24/7 assistance, announced alongside a revamped Gemini app that moves beyond chatbots into autonomous task execution.
- Smart Glasses — Google previewed AI-powered glasses launching this fall, with audio-driven interaction and contextual awareness.
- Search Redesigned — The New York Times reported that Google changed its search box for the first time in 25 years, powered entirely by AI results.
- Gmail Live — A new conversational interface for searching emails, coming to AI Plus and AI Pro subscribers.
- Genie World Model — Google’s world model can now simulate real streets using Street View data.
TechCrunch declared “Google just declared itself a contender in AI design at IO 2026,” while CNBC noted the company is making a concerted push to keep pace with OpenAI and Anthropic.
🚀 OpenAI Announces GPT-5.5
CNBC reports that OpenAI has announced GPT-5.5, its latest flagship model. This continues OpenAI’s rapid iteration cycle following GPT-5 earlier this year. No pricing or availability details have been released yet, but the announcement signals OpenAI’s determination to stay ahead of Anthropic’s Mythos and Google’s Gemini 3.5.
In other OpenAI news, the company partnered with Dell Technologies to bring Codex to hybrid and on-premises enterprise environments, and Sam Altman won the Elon Musk lawsuit, clearing the way for OpenAI’s trillion-dollar ambitions.
🚀 Meta Abandons Open-Source Llama for Proprietary Muse Spark
In a major strategic pivot, Meta is reportedly abandoning its open-source Llama strategy in favor of a new proprietary model stack called Muse Spark. Forbes reports that Muse Spark is Meta’s rebuilt AI architecture following what it considers Llama’s underwhelming performance in the competitive landscape.
The New Stack notes this is a significant reversal for Meta, which had been the leading advocate for open-weight AI models. The shift to closed-source raises questions about the future of open-source AI and Meta’s role in the ecosystem.
🤗 Trending on Hugging Face
Hugging Face saw mostly quiet activity today, with the standout being:
- elizaos/eliza-1 — 50K+ downloads, a Qwen-based GGUF/ONNX model tagged for the Eliza agent framework. One of the few models with real adoption metrics on today’s feed.
🚀 New Models on OpenRouter
- Google: Gemini 3.5 Flash — 1,048,576 token context, priced at $1.5/M tokens (prompt). This is the most significant OpenRouter addition today, giving developers access to Google’s latest frontier model at competitive pricing.
⭐ GitHub Trending: AI Edition
| Repo | Stars | Language | Description |
|---|---|---|---|
| Doorman11991/smallcode | 706★ | JavaScript | AI coding agent optimized for small LLMs; 87% benchmark with 4B-active model |
| mikesheehan54/Claude-Code-Design-AI | 451★ | TypeScript | AI UI/UX architect: screenshot to React, Figma components, Tailwind CSS generator |
smallcode is the standout today — an AI coding agent achieving 87% on benchmarks while using only a 4B-active parameter model, making it highly relevant for the local/edge AI movement.
💡 Key Trends
- Safety over capability: Anthropic withholding Mythos marks a turning point — frontier labs are now making market decisions based on risk assessment, not just capability benchmarks.
- Google’s agentic pivot: I/O 2026 made clear Google is betting everything on autonomous AI agents, not chatbots. Gemini Spark, Gmail Live, and the redesigned search all point toward proactive, task-executing AI.
- Open-source retreat: Meta’s shift from open-source Llama to proprietary Muse Spark signals that even the biggest open-source champions are reconsidering as the competitive pressure intensifies.
- Cost compression: Gemini 3.5 Flash at $1.5/M tokens with 1M context continues the trend of dramatically lowering cost per token while increasing context windows.