Apple AI Releases CLaRa | OpenAGI Team Launches Lux (Computer Use Model)..

Here is your today’s AI Dev Brief from Marktechpost, covering core research, models, infrastructure tools, and applied updates for AI developers and researchers.

Apple Researchers Release CLaRa-7B, a continuous latent reasoning framework that replaces raw documents with learned memory tokens and unifies retrieval and generation in a shared embedding space. A Mistral-7B backbone with LoRA adapters and SCP pretraining on ≈2M Wikipedia passages delivers 4x–128x semantic compression while improving average F1 over LLMLingua-2 by up to 17.31 points in Oracle settings and even outperforming BGE + full-text RAG, reaching 96.21 Recall@5 and 75 F1 on Natural Questions and HotpotQA at 4x compression Read the full insights/article here.

MiniMax has opened applications for its Developer Ambassador Program, aimed at independent ML and LLM developers who are already building with MiniMax models. Ambassadors get access to upgraded or free plans, early access to new releases, direct channels to the product and R&D teams, and visibility for their work through the MiniMax community and events Check out the details.

Lux is OpenAGI’s new foundation computer use model that runs real desktops and browsers, scores 83.6 percent on Online Mind2Web versus 69.0 for Gemini CUA, 61.3 for OpenAI Operator and 61.0 for Claude Sonnet-4, delivers about 1 second per step at roughly 10 times lower token cost, and is trained on large scale OS level trajectories collected with the open source OSGym engine that can run more than 1,000 replicas and generate more than 1,420 multi turn trajectories per minute. Read the full insights/article here.

Marktechpost has launched AIResearchCharts.com, a live analytics platform that aggregates and visualizes data from 5,000+ NeurIPS 2025 accepted papers in an interactive dashboard-style interface. The site turns conference accepted papers/data into queryable charts, enabling users to explore trends in research topics, track prolific institutions and authors, and map how work is distributed across subfields and geographies. By replacing manual spreadsheet analysis with filtered views and saved paper collections, the platform is designed to make questions about “who is doing what, and where the field is moving” answerable in a few clicks instead of hours of ad-hoc digging. Check out this Open Tool here.

▶ [Open Source] Rogue: An Open-Source AI Agent Evaluator worth trying Codes & Examples

▶ How to Design a Fully Local Multi-Agent Orchestration System Using TinyLlama for Intelligent Task Decomposition and Autonomous Collaboration Codes Tutorial

▶ How to Build a Meta-Cognitive AI Agent That Dynamically Adjusts Its Own Reasoning Depth for Efficient Problem Solving Codes Tutorial

▶ A Coding Guide to Design an Agentic AI System Using a Control-Plane Architecture for Safe, Modular, and Scalable Tool-Driven Reasoning Workflows Codes Tutorial

▶ A Coding Implementation for an Agentic AI Framework that Performs Literature Analysis, Hypothesis Generation, Experimental Planning, Simulation, and Scientific Reporting Codes Tutorial

100+ more open codes/notebooks here ➡️

For Sponsorship/Promotion, Please reach out at [email protected]