Perplexity Assistant released on iOS!

The revamped AI search app can now take users' commands and execute agentic actions

It can help with things like accessing the web, playing media, drafting emails, scheduling meetings, and more

ByteDance's Dreamina AI officially launched Seedream 3.0, a new text-to-image model

—Produces 2K cinematic visuals with realistic textures
—Delivers precise typography
— #2 on Artificial Analysis’ Image Arena behind only GPT-4o

Alibaba’s Qwen just unveiled Qwen3: a family of eight open models ranging from 600M to 235B params.

— Flagship version rivals OpenAI o1 & DeepSeek-R1
— Hybrid “thinking” mode in all models
— Boosted coding + agent performance
— Supports 119 languages

OpenAI enhanced ChatGPT Search with new features, including:

—An improved shopping UX with better product suggestions, reviews, and pricing comparisons
—Better citations with highlights
—Trending & autocomplete searches
—Searches via ChatGPT on WhatsApp

Luma Labs released Ray2 Camera Concepts in the API

The move allows developers to integrate the model’s advanced video controls (for cinematic framing and dynamic movement) into their applications

Higgsfield AI introduced Iconic Scenes!

The feature re-creates famous movie scenes with a different subject using just a single selfie from the user

Available to try on the Higgsfield AI website

San Francisco-based P-1 AI emerged from stealth with $23M in seed funding

The company is building "Archie," an engineering-focused AI agent aimed at automating cognitive engineering tasks

Figure AI is said to be partnering with the United Parcel Service to deploy its humanoids in shipping and logistics

This comes as CEO Brett Adcock takes legal action against Fortune over an article that alleges Figure overstated its robots' capabilities

Baidu dropped faster and cheaper ERNIE 4.5 Turbo and X1 Turbo models

4.5 Turbo, with upgraded multimodality, costs 11c and 44c per million I/O tokens (0.2% of GPT-4.5).

X1 Turbo comes at 14c/55c per million I/O tokens, topping DeepSeek R1 and V3

DeepMind built an AI that replicates how a fruit fly walks, flies, and behaves — using MuJoCo, its open physics simulator

The work will help scientists understand what drives specific behaviors in the fly, finding links that labs can't always measure

Listen Labs raised $27M from Sequoia for an AI-powered customer researcher

Their system runs thousands of interviews in parallel to uncover what users want and why—helping teams guide their product/sames pipelines

Already in use by Microsoft and Canva!

Elon Musk just confirmed that xAI will launch Grok 3.5 next week

He said that the upcoming version will reason from 'first principles' and provide answers that don't exist on the internet

However, it's only coming as an early beta to SuperGrok users!

Runway launched Gen-4 References to its paid plans

The tool allows users to use photos, images, 3D models, or selfies to place a character into any scene with consistency

Midjourney dropped new updates:

—Improved V7 model with enhanced image quality and coherence
—A new, easy-to-use lightbox editor
—A new experimental aesthetics parameter to play with details (goes from 0 to 100, with 0 being the default)

JetBrains just open-sourced Mellum, it's 4B param coding AI

First announced last year, the model is being touted as a 'focal' product designed specifically for code completion rather than chasing multiple capabilities

Available under Apache 2.0

Amazon's Nova Premier is now available on Bedrock

It's the company's most capable teacher model for creating custom distilled models for tasks like RAG, function calling, and agentic coding

Also includes a 1M-token context window

At LlamaCon 2025, Meta announced:

—Standalone Meta AI app with a social 'discover' feed to take on ChatGPT
—Llama API free preview
—Lama Guard 4 (12B), LlamaFirewall, and Prompt Guard
—Colab with Groq and Cerebras for faster inference

NVIDIA's CEO, Jensen Huang, shared his thoughts on the AI race with China, saying:

—The country has great capability and is 'right behind' the U.S
—Huawei has everything to advance AI (in terms of infra)
—It's a long-term, infinite competition

Google just announced the expansion of NotebookLM's podcast-generating Audio Overviews feature

It now supports over 50 languages, including Spanish and Hindi, for easy creation of multilingual content

Nvidia just open-sourced Describe Anything!

It's a new method to generate detailed descriptions for specific areas in images and videos

All you have to do is mark the region with points, boxes, scribbles, or masks!

As part of its Spring 2025 update, Yelp will test an AI-powered feature that lets restaurants and businesses handle customer calls using virtual agents.

This will help businesses operate even when their support staff is busy or unavailable