⏰ OpenRouter Limited-Time Free Models

Free inference remains available on Owl Alpha (1.05M context), NVIDIA Nemotron 3 Super (120B param MoE, 1M context), and Poolside Laguna M.1 (coding-optimized, 128K context). OpenRouter continues to expand free model capacity through new provider partnerships and direct cost coverage.

🤗 Trending on Hugging Face

This week’s HuggingFace trending slate features:

  • empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF (908K downloads) — a GGUF-quantized reasoning blend based on Qwen3.5, gaining rapid traction in the local-inference community.
  • deepreinforce-ai/Ornith-1.0-35B-GGUF (124K downloads) — another high-profile GGUF release for local deployment.
  • caiqi/difficulty-qwen2.5-0.5b-run-0623-maxlen3072 — a difficulty-prediction fine-tune with Safetensors support.
  • truongnt/hybridvla-checkpoints — vision-language-action model checkpoints for robotics.
  • SiddharthaChekuri/Nemotron-Super-49B-aie-v18-20k-ckpt100 — NVIDIA Nemotron NAS variant, 49B parameters with custom code.
  • kiikiik/gemma3-27b-gsm-sft-3epoch — Gemma 3 fine-tuned on GSM math reasoning (Apache 2.0).

Broader June trends: DeepSeek V4.1 Flash remains the top trending text-generation model; Qwen 3.7 and Gemma 4 (31B Dense) hold #2 and #3. Chinese open-weight models now occupy five of the top ten slots — the highest concentration on record.

⭐ GitHub Trending: AI Edition

  • winsznx/theeleven ⭐715 — Eleven autonomous AI agents that open live football prop markets on X Layer via a custom Uniswap v4 hook and gas-optimized smart contracts.
  • benchflow-ai/awesome-evals ⭐591 — A curated, no-BS library of the best resources for building and evaluating AI agents — papers, blog posts, and frameworks.
  • Pluviobyte/video-production-skills ⭐456 — Reusable AI video production skills library covering creation, motion design, openers, and QA workflows.
  • tutusagi/ai-fishing-game ⭐313 — A deterministic text-based fishing game designed for AI companions — single-file, zero dependencies, Python.
  • eli-labz/Godcoder ⭐251 — A local-first, open-source coding agent for desktop. Bring your own LLM key; code stays on your machine.

💡 Key Trends

  1. The $84.75B Infrastructure Bet — Alphabet closed the largest equity financing in corporate history for AI compute infrastructure. Berkshire Hathaway contributed $10B (its largest single tech investment). Pichai noted that compute demand exceeds supply; 2026 capex guidance is $180–$190B with 2027 projected even higher.

  2. Google’s Agentic Execution Emergency — Brin’s internal memo named “agentic execution” (not model quality) as the critical gap vs. Anthropic. Google codes ~50% with AI vs. Anthropic’s ~100%. Six researchers have left in five months for Meta, OpenAI, and Anthropic. The newly formed coding strike team has added a midtraining phase between pretraining and post-training.

  3. GitHub Copilot Metered Billing Shock — The first full cycle of usage-based billing saw enterprise bills jump 10–50× (e.g., $29 → $750, $50 → $3,000). One agentic session costs $30–$40. This marks the end of the flat-rate AI era — Cursor and Windsurf already use metered billing. A two-tier structure ($20 Pro / ~$200 Power) is emerging across the industry.

  4. Qualcomm’s $3.92B Software Infrastructure Move — Qualcomm acquired Modular (Mojo language + MAX platform) to give its inference chips a software layer competitive with Nvidia CUDA. This positions Qualcomm as a serious enterprise AI hardware contender in H2 2026.

  5. AI Workforce Reshaping Accelerates — Anthropic’s Economic Index found 35% of users expect AI to perform most tasks within 12 months. AI has driven 88,000 US job cuts in 2026. The RAISE US nonprofit ($500M+ secured) launched to retrain American workers, backed by Amazon, Anthropic, Microsoft, and OpenAI Foundation.

Today’s Snapshot

  • HuggingFace models trending: 10 new entries today
  • OpenRouter new models: 0 (no new listings today)
  • GitHub trending repos: 5 (total ⭐2,326)
  • Papers: 0
  • Limited-free models: 3 (stable lineup)
  • Total items tracked: 15