🤖AI Agents Weekly: Claude Opus 4.5, OmniScientist, FLUX.2, General Agentic Memory
Claude Opus 4.5, OmniScientist, FLUX.2, General Agentic Memory
In today’s issue:
OmniScientist: AI scientists framework
Claude Opus 4.5: Anthropic’s flagship model
Fara-7B: Efficient computer use agent
MiniMax-M2 Deep Research agent
General Agentic Memory (GAM)
Gemini 3 API updates
FLUX.2 image generation
Exa API 2.1 search tiers
MCP Apps for interactive UIs
Event-Centric Memory for agents
Agent-as-a-Graph retrieval
Top AI dev news and papers
Top Stories
OmniScientist
Researchers introduce OmniScientist, an end-to-end framework for building AI scientists capable of autonomously conducting research across the entire scientific lifecycle. The system integrates specialized agents with a massive knowledge foundation spanning 250M+ papers and 100M+ scientific entities, establishing a collaborative ecosystem where human and AI scientists can work together on real-world discoveries.
Complete scientific workflow – OmniScientist covers five core stages: literature review using retrieval and graph-based discovery, research ideation powered by 10M+ idea seeds, experiment automation through code generation and lab integration, scientific writing with structured drafting, and paper review via multi-agent critique and refinement.
Omni Scientific Protocol (OSP) – A structured communication standard that enables seamless collaboration between humans and AI agents. OSP defines roles, task formats, and interaction patterns, allowing researchers to delegate subtasks, review outputs, and iteratively refine results while maintaining scientific rigor and reproducibility.
ScienceArena evaluation platform – A comprehensive benchmark suite with 1,500+ expert-verified tasks across multiple disciplines, measuring AI scientists on retrieval accuracy, ideation novelty, experimental correctness, writing quality, and review consistency. The platform tracks contribution provenance, enabling transparent credit assignment in human-AI co-authorship.
Data foundation and real-world deployment – Built on OpenAlex metadata, arXiv full-texts, and scientific knowledge graphs, OmniScientist demonstrates practical applications in materials science, drug discovery, and computational biology. The framework supports continuous learning, with AI agents improving through feedback loops and community contributions.