⏰ OpenRouter Limited-Time Free Models

Model Context Pricing Expires
Poolside: Laguna XS.2 (Free) 262K tokens $0/M tokens Jul 9 (5 days left)
Poolside: Laguna XS.2 262K tokens $0.10/M input → $0.20/M output Jul 9 (5 days left)

The Poolside Laguna XS models are wrapping up their limited-time free window with just 5 days remaining. Poolside’s newer Laguna XS 2.1 is also available as a paid option at $0.06/M input, offering incremental improvements for software development tasks.

🤗 Trending on Hugging Face

This wave sees mostly niche and domain-specific uploads rather than a breakout hit. Notable entries include:

  • sundaycoil/config-watcher — A configuration monitoring utility, suggesting growing interest in agent tooling infrastructure on HF.
  • mradermacher/Goetia-26B-A4B-v1.4-i1-GGUF — A 26B-parameter MoE merge with GGUF quantization, built using mergekit and MoE-Della techniques. Demonstrates the community’s continued push toward efficient local inference.
  • gmxch/fa_icon — A Vision Transformer model released under MIT license, with safetensors format.
  • NamanSoni78/image-to-ghibli — A Ghibli-style image translation model, riding the ongoing anime aesthetic wave.
  • seed429/trained — A DPO-trained model built with TRL (Transformer Reinforcement Learning), part of the growing RLHF-from-scratch movement on HF.

⭐ GitHub Trending: AI Edition

Repo Description Stars
Kulaxyz/self-learning-skills AGENTS.md-driven self-improvement workflow for AI coding agents — learn from past mistakes autonomously ⭐ 974
TianhangZhuzth/Fundamental-Ava Framework for building autonomous, collaborative, and socially intelligent digital human agents ⭐ 759
HUANGCHIHHUNGLeo/claude-real-video Let Claude (or any LLM) actually watch a video — scene-aware, deduplicated frames + transcript ⭐ 413
JasonLiu0826/ackem Local-first AI desktop companion — privacy-preserving, with memory/emotion/extension support ⭐ 310
lingbol088-spec/reverse-flow-skill Local CTF reverse engineering workflow for AI agents — analysis to decompilation pipeline ⭐ 301

💡 Key Trends

  1. The Chip Independence Race Accelerates. Anthropic’s reported talks with Samsung for custom AI chips — directly targeting OpenAI’s Jalapeño silicon strategy — mark a new phase in AI hardware competition. AI labs are no longer content to rent Nvidia’s ecosystem; they want their own silicon, their own supply chains, and their own moats. Samsung, with its memory and fabrication expertise, could become the foundry partner of choice for the post-Nvidia era.

  2. Governments Move Toward AI Release Frameworks. The US government is nearing a voluntary deal with AI companies on how frontier models get released — an announcement could come as soon as next week. Meanwhile, Meta is quietly building a cloud business to monetize its spare AI compute, a defensive hedge against the $37B+ annual CapEx it’s pouring into infrastructure. The BIS warns that the AI spending boom may be outrunning its returns.

  3. Agent Infrastructure Matures Across the Stack. From multi-agent hierarchical code summarization (Agent4cs) to automated federated learning research and time-series foundation models (t0-alpha), the agent ecosystem is expanding beyond chat. The GitHub trending list confirms this shift: self-improving coding agents, video-aware LLM tools, and local-first AI companions all point to a market where agents, not raw models, are the differentiator.

  4. OpenRouter’s Model Menu Expands. Claude Sonnet 5 arrives at $2.00/M input — near-flagship performance at a competitive price tier. Gemini 3.1 Flash Lite Image brings multimodal to the budget segment. Poolside’s Laguna 2.1 series continues its push into specialized software development inference. The platform now spans from free tier to premium flagship, giving developers unprecedented choice.

Today’s Snapshot

  • 10 new Hugging Face models (MoE merges, DPO training, niche domain models)
  • 4 new OpenRouter models (Claude Sonnet 5, Poolside Laguna XS 2.1, Gemini 3.1 Flash Lite Image)
  • 5 trending GitHub repos (agent tooling, AI companions, video understanding)
  • 0 new papers found
  • 2 limited-time free models ending soon