⏰ OpenRouter Limited-Time Free Models

Several capable models remain free on OpenRouter through mid-July. NVIDIA Llama 3.3 Nemotron Super 49B V1.5 ($0.40/M tokens) and Meta Llama 3.2 11B Vision Instruct ($0.35/M tokens) are free until July 17. Qwen3 Next 80B A3B Instruct and Qwen3 Coder 480B A35B — both with 1M+ context — are free through July 19. Tencent Hy3 (262K context) is free until July 21. Nous Hermes 3 405B Instruct and Meta Llama 3.3 70B Instruct remain free until July 19.

🤗 Trending on Hugging Face

New uploads this cycle include KordAI/Kord-Translate-ENTH-V2-8B, a Qwen3-based translation model for English-Thai, and brownianeffect/Flux-2-Klein-brownian (Apache 2.0 licensed). Several GGUF quantized models were also uploaded, including mradermacher/Dark-Pane-GGUF — a conversational model packaged for local inference with llama.cpp.

⭐ GitHub Trending: AI Edition

simonlin1212/Vibe-Research — ⭐ 708 — Personal trading research agent covering A-shares, US stocks, and HK stocks with daily review, news radar, and portfolio tracking. Built in TypeScript.

nexu-io/motion-anything — ⭐ 380 — An open-source, chat-native motion engine that translates natural language descriptions into animations. JavaScript.

Doriandarko/texts-to-transformer — ⭐ 336 — Train a tiny Transformer from scratch on your iMessage history, entirely on your Mac. Python.

ronak-create/FableCut — ⭐ 325 — Zero-dependency browser video editor that AI agents can drive via JSON timeline, MCP, and REST. JavaScript.

yusukebe/ax — ⭐ 289 — Described as “the AI-era curl,” a TypeScript tool for CLI-based AI interactions.

💡 Key Trends

  1. OpenAI’s Platform Expansion Accelerates — The simultaneous launch of GPT 5.6 (in three tiers: Luna, Terra, Sol) and ChatGPT Work marks a major push from conversational AI to autonomous project completion. The GPT-Live voice model family further extends OpenAI’s reach into real-time voice interaction, delegating reasoning to GPT-5.5 under the hood.

  2. Agentic AI Goes Mainstream — ChatGPT Work, Anthropic’s Claude Cowork for mobile/web, and DeepSeek’s chip plans all point to an industry-wide shift: AI is no longer just answering questions but executing multi-step workflows autonomously. Claude Cowork on phones signals the mobile agent race is heating up.

  3. Infrastructure Demand Strains Sustainability — Microsoft’s greenhouse gas emissions rose 27% to 21.1M metric tons driven by AI data center expansion. Meanwhile, HFCL is scaling optical fiber production to meet hyperscaler demand, and NVIDIA’s Nemotron model claims 2.03x server throughput improvement — a reminder that efficiency gains and sustainability are on a collision course.

  4. Multimodal and Sensory AI Advances — Meta’s “Super Sensing” AI glasses (always-on visual/audio awareness) and Muse image generation for Instagram/WhatsApp, plus ByteDance’s Seedream 5.0 Pro with design-focused AI, show the race toward ambient AI that sees, hears, and generates seamlessly.

Today’s Snapshot

  • New OpenRouter models: 8 (GPT 5.6 family × 6 + Grok 4.5 × 2)
  • Trending GitHub repos: 5
  • Hugging Face uploads: 10
  • Limited-time free models: 10