๐GITHUB WATCH: "Reverse Engineer Anything" passed 342,000 stars, just as Claude and Codex began blocking users reverse-engineering their own devices. Open tools, closed guardrails. #OpenWeightUnderground#AI#GenAI
๐FOR BUILDERS: Google MaxText now supports long- context training up to 14M tokens on TPUs.
Open-source, JAX-based, and now very long.
Read the docs before you burn the budget. ๐ฌ
#OpenWeightUnderground#OpenSourceAI#AI#GenAI
๐ ODD BUT USEFUL: Sarvam Prism
Open-weight decision model (LoRA on Sarvam 105B) that scores predefined options in a single forward pass, with no token-by-token generation.
Sometimes the best answer is a pick, not a paragraph. ๐ฏ
#OpenWeightUnderground
๐ HIDDEN GEM: NorthCUA
Open-source agent suite for desktop & web tasks, built on Qwen3.5: Nano (9B), Mini (35B), Pro (397B).
Reported: Pro beats Qwen3.8-Max & Kimi-K3 on OSWorld-Verified.
Test on your own apps first. ๐ฅ๏ธ
#OpenWeightUnderground
A powerful model doesnโt always need a closed door. DeepSeek-V3.2 brings a mixture-of-experts approach to the open-model conversation, with capabilities aimed at reasoning, coding and more.
#OpenWeightUnderground#OpenSourceAI#LLMs
Have you explored Meta's Llama 3.3 70B yet? A powerful open-weight model for builders, developers and AI enthusiasts. Would you give it a try?
#OpenWeightUnder#Llama3#OpenSourceAI
One model. Two ways to work. Meet Mage-VL-Cowork, combining screen interaction with code execution. Small model, interesting possibilities!
#OpenWeightUnder#AI#GenAI
Hidden gems in open AI are worth discovering. Meet Bria FIBO 2.0 Dev, where image generation and editing come together in one visual workflow.
#AI#GenAI#EnterpriseAI