πŸ€– Hugging Face New Models

This week’s Hugging Face releases showcase remarkable diversity in model development:

  • evolai-d (Qwen3-based) β€” 1,893 downloads, featuring safetensors format
  • evolai-mamba2-c β€” 1,306 downloads, Mamba2 architecture for efficient long-sequence processing
  • tbuckley/Qwen2.5-7B-Instruct_risky-financial-advice_kl-narrow-kl3e3-seed1 β€” specialized financial advice model with safety training
  • koyelog/MediMind-411M β€” medical LLM for targeted healthcare applications
  • JWei05/gemma3-4b-pt-sft-distill-from-12b-rl-step20-seed43 β€” distilled Gemma-3 4B model
  • sara0123456789/urdu-gec-mt5-A3 β€” Urdu grammar correction using mT5 architecture
  • SOTAagi2030/MyAwesomeModel-TestRepo β€” BERT-based feature extraction model
  • aioaneid/nanochat_n_layer_12_seq_len_1024_n_embd_1024 β€” compact chat model with MIT license

πŸ’° OpenRouter Models β€” Ultra-Low Cost Highlights

OpenRouter continues to democratize access to frontier models with aggressively priced options:

Model Context Length Prompt Price
inclusionAI: Ring-2.6-1T (free) 262K $0.00
Google: Gemini 3.1 Flash Lite 1M $0.00000025/1M tokens

The free Ring-2.6-1T model offers 262K context at zero cost, while Gemini 3.1 Flash Lite delivers 1M token context for mere fractions of a cent.

⏰ Limited-Time Free Models

Several models are currently available for free or at ultra-low cost:

  • Anthropic: Claude 3.7 Sonnet (expires May 11) β€” $0.000003 prompt, 200K context
  • Anthropic: Claude 3.7 Sonnet (thinking) (expires May 11) β€” extended reasoning mode
  • Z.ai: GLM 4.6 (expires May 14) β€” $0.00000039 prompt, 204K context
  • MoonshotAI: Kimi K2 0905 (expires May 14) β€” $0.0000004 prompt, 262K context
  • xAI: Grok 4.1 Fast (expires May 15) β€” $0.0000002 prompt, 2M context β€” standout offer!
  • xAI: Grok 4 Fast (expires May 15) β€” same pricing, 2M context
  • xAI: Grok 4 (expires May 15) β€” full Grok-4 at $0.000003 prompt

xAI dominates the limited-free tier with Grok models offering industry-leading 2 million token context windows.

🌟 GitHub Trending Repos

1. strukto-ai/mirage β€” 1,619 ⭐

“A Unified Virtual Filesystem For AI Agents” TypeScript-based project enabling AI agents to interact with multiple data sources through a unified virtual filesystem interface.

2. yaojingang/yao-open-prompts β€” 1,481 ⭐

Chinese AI Prompt Library covering work, study, content creation, marketing, and daily life scenarios β€” a comprehensive resource for Chinese-language AI practitioners.

3. lightseekorg/tokenspeed β€” 854 ⭐

“Speed-of-light LLM inference engine” β€” pushing the boundaries of inference performance optimization.

4. raisyanyahya/how-to-train-your-gpt β€” 779 ⭐

“Build a modern LLM from scratch” β€” every line commented and explained, perfect for learning transformer architecture from the ground up.

5. WenyuChiou/awesome-agentic-ai-zh β€” 521 ⭐

AI Agent δΈ­ζ–‡ε­ΈηΏ’εœ°εœ– β€” trilingual (Traditional Chinese/Simplified Chinese/English) structured learning path for AI agents, with exercises and required readings at each stage.

πŸ”‘ Key Trends This Week

  1. Local AI Acceleration: Mamba2 architecture models and compact nanochat variants signal continued push toward efficient local deployment.

  2. 2M Context Race: xAI’s Grok 4.1 Fast with 2 million token context at $0.0000002/prompt sets a new value benchmark.

  3. Multilingual Focus: Urdu GEC model and Chinese AI agent learning resources show growing non-English AI ecosystem.

  4. Agent Infrastructure: Mirage’s virtual filesystem approach to AI agent data access reflects maturing agent toolchains.

Collected: 2026-05-09 | HF: 10 models | OR: 2 models | GitHub: 5 repos | Total items: 27