Editor’s Pick
You should not miss this one
[Open Reasoning Model] Ant Group Releases Ling 2.0: A Reasoning-First MoE Language Model Series Built on the Principle that Each Activation Enhances Reasoning Capability. How do you build a language model that grows in capacity but keeps the computation for each token almost unchanged? The Inclusion AI team from the Ant Group is pushing sparse large models in a methodical way by releasing Ling 2.0. Ling 2.0 is a reasoning based language model family built on the idea that each activation should translate directly into stronger reasoning behavior. It is one of the latest approaches that shows how to keep activation small while moving from 16B to 1T without rewriting the recipe. The series has three versions, Ling mini 2.0 at 16B total with 1.4B activated, Ling flash 2.0 in the 100B class with 6.1B activated, and Ling 1T with 1T total and about 50B active per token......
AI Dev and Latest Releases
[IBM’s Open Small Model] IBM AI Team Releases Granite 4.0 Nano Series: Compact and Open-Source Small Models Built for AI at the Edge. Small models are often blocked by poor instruction tuning, weak tool use formats, and missing governance. IBM AI team released Granite 4.0 Nano, a small model family that targets local and edge inference with enterprise controls and open licensing. The family includes 8 models in two sizes, 350M and about 1B, with both hybrid SSM and transformer variants, each in base and instruct. Granite 4.0 Nano series models are released under an Apache 2.0 license with native architecture support on popular runtimes like vLLM, llama.cpp, and MLX.
[New Open Source MoE Model] MiniMax releases MiniMax M2: A Mini Model Built for Max Coding and Agentic Workflows at 8% Claude Sonnet Price and ~2x Faster. M2 is an open-weight MIT-licensed MoE for coding and agents, listed at 229B total parameters with about 10B active per token. It ships on Hugging Face with deployment notes for vLLM and SGLang and requires preserving <think> blocks in history. Reported evaluations target agent and code tasks, not only static QA. The API lists $0.30 per 1M input tokens and $1.20 per 1M output tokens, with a limited free access window. It also claim cost far below Claude Sonnet and about 2 times the speed.
[Open Small Model] Liquid AI Releases LFM2-ColBERT-350M: A New Small Model that brings Late Interaction Retrieval to Multilingual and Cross-Lingual RAG. LFM2-ColBERT-350M is a late interaction retriever that encodes queries and documents separately, then applies MaxSim over token embeddings at query time, which preserves token level interactions and allows precomputed document embeddings for scale. It targets multilingual and cross lingual retrieval so teams can index documents once and answer queries in many languages with high accuracy. Liquid AI reports inference speed on par with models that are 2.3 times smaller, and provides a public demo and model card with usage details for PyLate and PLAID. This makes it a practical drop in component for multilingual RAG systems.
[Open Source Agentic] Microsoft Releases Agent Lightning: A New AI Framework that Enables Reinforcement Learning (RL)-based Training of LLMs for Any AI Agent. gent Lightning decouples agent execution from reinforcement learning, exposes a unified trace interface, and uses LightningRL to convert multi step trajectories into single turn training transitions with credit assignment and Automatic Intermediate Rewarding, enabling optimization of existing agents in LangChain, OpenAI Agents SDK, AutoGen, and more with minimal code change, with reported gains on Spider, MuSiQue, and Calc X using Llama 3.2 3B Instruct.....
Project Notebooks/Tutorials
▶ How to Build a Fully Functional Computer-Use Agent that Thinks, Plans, and Executes Virtual Actions Using Local AI Models Codes Tutorial
▶ How to Use python-A2A to Create and Connect Financial Agents with Google’s Agent-to-Agent (A2A) Protocol Notebook-inflation_agent.py Notebook-network.ipynb Notebook-emi_agent.py Tutorial
▶ How to Build Ethically Aligned Autonomous Agents through Value-Guided Reasoning and Self-Correcting Decision-Making Using Open-Source Models Codes Tutorial
▶ How to Build, Train, and Compare Multiple Reinforcement Learning Agents in a Custom Trading Environment Using Stable-Baselines3 Codes Tutorial
▶ How to Build an Agentic Decision-Tree RAG System with Intelligent Query Routing, Self-Checking, and Iterative Refinement? Codes Tutorial
Our Latest: ‘GPU Leaderboard ‘
Currently in Beta release
Marktechpost introduces GPULeaderboard.com (beta release). It is a comprehensive comparison platform for AI researchers and developers seeking optimal GPU cloud solutions. Compare real-time pricing across 50+ providers including AWS, Google Cloud, RunPod, and Vast. ai for NVIDIA H200, B200, H100, A100 GPUs. Features include cost calculators, performance benchmarks, provider rankings, and deal-finding tools. Essential for finding the most cost-effective GPU instances for machine learning workloads, training large models, and AI development projects worldwide…..
