⏰ OpenRouter Limited-Time Free Models
| Model | Expires | Context | Input/Output Price |
|---|---|---|---|
| Qwen: Qwen3 Next 80B A3B Instruct (free) | Jul 19 | 262K | Free |
| Qwen: Qwen3 Coder 480B A35B (free) | Jul 19 | 1M | Free |
| Venice: Uncensored (free) | Jul 19 | 33K | Free |
| Meta: Llama 3.3 70B Instruct (free) | Jul 19 | 131K | Free |
| Meta: Llama 3.2 3B Instruct (free) | Jul 19 | 131K | Free |
| Nous: Hermes 3 405B Instruct (free) | Jul 19 | 131K | Free |
| Tencent: Hy3 (free) | Jul 21 | 262K | Free |
| Poolside: Laguna M.1 (free) | Jul 28 | 262K | Free |
| Poolside: Laguna M.1 | Jul 28 | 262K | $0.20/M input, $0.40/M output |
| OpenAI: GPT-5.3 Chat | Aug 10 | 128K | $1.75/M input, $14/M output |
Several free models expire Jul 19–21. The Tencent Hy3 free tier, Poolside Laguna M.1, and GPT-5.3 Chat remain available through late July and August.
🆕 New Models on OpenRouter
MoonshotAI: Kimi K3
MoonshotAI has released Kimi K3 on OpenRouter at $3/M input and $15/M output tokens with a 1M context window. This is MoonshotAI’s new flagship model, positioned competitively against GPT-5.6 Terra and Claude Sonnet 5 in both pricing and capability.
Meta: Muse Spark 1.1
Meta has updated its Muse Spark line to version 1.1, priced at $1.25/M input and $4.25/M output with 1M context. The update improves reasoning and coding performance while maintaining cost efficiency.
📰 Top AI News
Google Delays Gemini 3.5 Pro — Coding Shortfalls Cited
Google has paused the public release of Gemini 3.5 Pro, originally unveiled at Google I/O in May. Internal reviews found that the model’s coding capabilities did not meet quality standards. The delay comes at a critical time as competitors OpenAI, Meta, Anthropic, and xAI all prepare new coding-focused models.
EU Demands Google Open Android to Rival AI Assistants
The European Union has asked Google to open key Android features and data to competitors under the Digital Markets Act, aiming to give rival AI assistants (including OpenAI) broader access to the Android ecosystem. Google warns the changes could weaken privacy and security guardrails.
OpenAI’s GPT-Red: AI That Red-Teams AI
OpenAI has introduced GPT-Red, an AI system that automatically finds vulnerabilities in other AI models — performing prompt injection testing, jailbreak detection, and safety boundary probing before models are released to the public.
OpenAI’s Miles Wang in Talks for $2B AI Drug Discovery Startup
Miles Wang is leaving OpenAI to start an AI-driven drug discovery company, reportedly valued at $2B. The startup would apply large language models to pharmaceutical research and development.
Mistral Vibe for Code Leads Multi-Agent Programming Benchmark
Mistral Vibe for Code has taken the top spot in a new multi-agent programming benchmark, outperforming Claude Code, Cursor, and OpenAI Codex in translating feature requests to working pull requests without human intervention.
OpenAI Targets 2027 for First Hardware: ChatGPT Speaker
OpenAI’s first hardware device will be a screenless ChatGPT smart speaker, targeted for a 2027 release. The device will be movable and voice-first, according to a Bloomberg report.
PrismML’s Bonsai 27B Runs Qwen3.6 on Laptops
PrismML has released Bonsai 27B, enabling Qwen3.6 to run on consumer laptops using 1-bit and ternary quantization. This marks a significant step in making frontier-capable models accessible without cloud GPUs.
DeepSeek Seeks More Capital Weeks After $7B Round
DeepSeek is in talks for additional funding just weeks after closing a $7B round, signaling aggressive expansion plans and ongoing compute infrastructure buildout.
Anthropic’s Claude for Teachers Vows Student Data Privacy
Anthropic has announced that Claude for Teachers will not train on student data, addressing a key concern in educational AI adoption.
Publishers Sue Google Over AI Book Training
Hachette, Cengage, Elsevier, and novelist Scott Turow have joined a class action against Google, alleging unauthorized use of copyrighted books for AI model training.
🤗 Hugging Face New Uploads
Notable uploads include Kepler-8B-Instruct GGUF (a conversational model by mradermacher), orqo-hype-v1 (CoreML/GGUF/peft compatible), and a Catalan language GPT-2 model for the NLD-Latn locale.
⭐ GitHub Trending: AI Edition
- PengZhang64/circuit-framework — Multi-agent LLM trading research system for quantitative finance. ⭐ 443
- Blueturboguy07/cue — Open-source macOS AI copilot that floats over your screen, sees/hears your meetings, and stays hidden. ⭐ 415
- QuantumByteOSS/quantumbyte — Open-source app builder engine — intent to working app. ⭐ 316
- nihe0909/xiaohongshu-ai-workbench — Free Codex Skills for Xiaohongshu (Little Red Book) social media operations. ⭐ 230
- barretlee/agent-pulse — Evidence-backed AI industry intelligence with daily data refreshes and weekly reports. ⭐ 202
💡 Key Trends
-
Regulatory Wave Intensifies — The EU’s DMA push against Google Android, OpenAI’s state-level regulation proposal, and publisher lawsuits against Google all point to accelerating AI governance activity. The debate is shifting from whether to regulate to who sets the rules.
-
Agentic Coding Race Heats Up — Mistral Vibe for Code topping the multi-agent benchmark, combined with Google’s Gemini 3.5 Pro delay over coding quality, shows that autonomous code generation is now the primary competitive battleground among frontier labs.
-
AI Hardware Diversifies — OpenAI’s ChatGPT speaker (2027) and PrismML’s laptop-run Qwen3.6 (1-bit quantization) represent two ends of the AI hardware spectrum: consumer appliances vs. on-device inference, both expanding the AI footprint beyond cloud APIs.
-
AI Talent Spins Off — Miles Wang leaving OpenAI for a $2B drug discovery startup and DeepSeek raising more capital weeks after $7B indicate that AI talent and capital are flowing into vertical applications, not just foundation model building.
Today’s Snapshot
- OpenRouter limited-free models: 10 active
- New models on OpenRouter: 2 (Kimi K3, Muse Spark 1.1)
- Hugging Face new uploads: 10
- GitHub trending AI repos: 5
- Top stories: Google Gemini 3.5 Pro delay, EU DMA Android order, OpenAI GPT-Red
Covering Jul 15–17, 2026. Sources: OpenRouter, Hugging Face, GitHub, Times of AI, AI Daily Post.