AI News — Thursday, March 26, 2026

10
OpenAI Introduces GPT-5.4 Mini and Nano Models

OpenAI has announced the release of GPT-5.4 mini and nano, new compact versions of their advanced language models, likely aimed at more efficient deployment and specific use cases.

OpenAI Blogproduct
9
Google Unveils TurboQuant AI Memory Compression Algorithm, Drawing 'Pied Piper' Comparisons

Google has introduced TurboQuant, a new AI memory compression algorithm, which the internet is humorously likening to the fictional 'Pied Piper' technology from Silicon Valley.

TechCrunchproduct
9
Anthropic Launches Cowork: A Claude Desktop Agent for File Interaction Without Coding

Anthropic has released Cowork, a new Claude desktop agent designed to interact directly with user files, enabling advanced AI assistance without requiring any coding knowledge.

VentureBeatproduct
8
Railway Secures $100 Million to Compete with AWS in AI-Native Cloud Infrastructure

Railway has raised $100 million in funding to develop AI-native cloud infrastructure, positioning itself as a significant challenger to established cloud providers like AWS.

VentureBeatindustry
8
Nous Research Releases NousCoder-14B, an Open-Source Coding Model

Nous Research has launched NousCoder-14B, an open-source coding model that enters the market at a time of high demand for AI-powered code generation and assistance.

VentureBeatopen-source
7
Google AI Blog Unveils Lyria 3, New Music Generation Model for Developers

Google has announced Lyria 3, its latest music generation model, making it available to developers for creating new audio experiences.

Google AI Blogproduct
7
OpenAI Launches Safety Bug Bounty Program to Enhance AI Security

OpenAI has initiated a Safety Bug Bounty program, inviting researchers and security experts to identify and report vulnerabilities in their AI systems to improve overall safety.

OpenAI Blogindustry
7
Deccan AI Raises $25M to Source Experts from India for AI Training, Competing with Mercor

Deccan AI has secured $25 million in funding to build an AI training workforce primarily from India, aiming to compete with companies like Mercor in the expert sourcing market.

TechCrunchindustry
7
MinerU-Diffusion: Rethinking Document OCR as Inverse Rendering via Diffusion Decoding

A new research paper proposes MinerU-Diffusion, a novel approach that reinterprets document OCR as an inverse rendering problem solved through diffusion decoding, potentially improving accuracy.

Hugging Faceresearch
6
WildWorld: A Large-Scale Dataset for Dynamic World Modeling in Generative ARPGs

Researchers introduce WildWorld, a substantial dataset designed for dynamic world modeling with actions and explicit states, aiming to advance generative capabilities in Action Role-Playing Games (ARPGs).

Hugging Faceresearch
6
SpecEyes: Accelerating Agentic Multimodal LLMs via Speculative Perception and Planning

SpecEyes presents a method to accelerate agentic multimodal LLMs by employing speculative perception and planning, enhancing their efficiency in complex tasks.

Hugging Faceresearch
6
Survey Explores Workflow Optimization for LLM Agents: From Static Templates to Dynamic Runtime Graphs

A new survey provides an overview of workflow optimization techniques for LLM agents, tracing the evolution from static templates to more dynamic runtime graph structures.

Hugging Faceresearch
6
The AI Skills Gap is Widening, with Power Users Gaining a Significant Advantage

An AI company reports that a growing AI skills gap is emerging, where 'power users' who effectively leverage AI tools are significantly outperforming others in the workforce.

TechCrunchindustry
5
your agent can think. it can't remember.

This article discusses the current limitations of AI agents, highlighting that while they possess reasoning capabilities, they often lack persistent memory, posing a challenge for complex, multi-step interactions.

Dev.toindustry
5
Lyria 3 Pro: Create Longer Tracks in More Google Products

Google expands its Lyria 3 music generation capabilities with a 'Pro' version, enabling users to create longer audio tracks within a wider range of Google products.

Google AI Blogproduct