Weekly AI #24

Welcome to Weekly AI #24
Alibaba has launched Qwen 3, a groundbreaking family of open-weight AI models, featuring dense and MoE architectures ranging from 0.6B to 235B parameters. The flagship Qwen3-235B-A22B rivals top-tier models like GPT-4o and Gemini 2.5 Pro in coding and reasoning, while smaller variants (e.g., Qwen3-4B) outperform predecessors like Qwen2.5-72B—showcasing dramatic efficiency gains.

Top 5 Tools

Create.ad - Create ads with zero effort

Lovart - AI design agent

AI Nail Designs Lab - Create Perfect Nail Designs, AI nail art, custom manicure, nail patterns generator

Imagine Create AI - Create stunning product and fashion photography with AI

HiDream-E1 - HiDream-E1 is an image editing model built on HiDream-I1

Top News / Shorts

Mon, Apr 28
- ChatGPT Search adds shopping, WhatsApp queries and trending features
- Windsurf upgrades free plan with 25 premium prompts and new logo
- Hinton opposes OpenAI restructuring, urges AGI safety protections
- Qafind Labs launches world's first Diffusion Language Model - ChatDLM
- Step1X-Edit launches open-source as AI image editor tool
- Genspark Super Agent adds one-click AI personalization
- Gamma adds AI image editing with animation features
- Luma AI launches Ray2 Camera API for cinematic framing control
- Higgsfield AI launches Iconic Scenes for movie role selfies
- HiDream open-sources E1 image editor with top benchmark scores
- Amazon-backed Glacier builds AI robots for smart recycling
- Suno teases upcoming smarter prompt capabilities
- P-1 AI raises $23M to build engineering AGI "Archie"
- Alibaba releases Qwen3-8B with 36T-token multilingual training
- Australian radio station uses AI host Thy for 6 months without listeners knowing
- PageOn AI 2.0 automates slide decks via conversational commands

Tue, Apr 29
- Meta's new AI app personalizes responses with social media data
- Prime Intellect completes world's first decentralized 32B AI training
- Meta launches Llama API preview for developers with free access
- WhatsApp to get encrypted AI features via Meta's Private Processing
- Microsoft Copilot adds WhatsApp image analysis feature
- Google debuts bite-sized language lessons via Gemini API
- NotebookLM expands beyond English with new audio capabilities
- Meta at LlamaCon unveils AI app, security tools & hardware deals
- Higgsfield AI launches beta API for cinematic video generation
- Tempolor AI upgrades music creation with text-to-soundtrack tools
- HeyGen launches Voice Mirroring for ultra-realistic AI avatars
- Grok 3.5 beta launches next week with first-principles reasoning
- Figure AI partners with UPS for humanoid logistics robots
- Qwen3 now runs locally via LM Studio in GGUF and MLX
- Alibaba's Qwen3-235B rivals top AI models in benchmarks

Wed, Apr 30
- Runway launches Gen-4 References for consistent AI-generated characters
- Suno teases upcoming "Cover with Personas" feature
- Bee launches as the first wearable AI companion in US
- DeepMind's SAS prompt boosts robot skills using ping-pong data
- Aero-1-Audio achieves SOTA results on just 16 H100 GPUs
- Krea AI Enhancer adds 22K upscaling via Topaz integration
- Alibaba releases lightweight Qwen2.5-Omni-3B with 50% VRAM savings
- Affogato AI launches realistic talking video from single images
- JetBrains open-sources Mellum 4B for focused code completion
- Recraft adds OpenAI-powered image editing and upscaling
- DeepSeek releases Prover-V2 for advanced math theorem proving
- Xiaomi launches MiMo-7B-RL model outperforming OpenAI o1-mini
- OpenAI reverts GPT-4o to fix excessive user flattery issues
- Google Cloud unveils Gemini+Imagen+Veo AI video creator on Vertex
- Poe adds Alibaba's powerful Qwen3 AI models
- Google AI trains sensor model from 165K users 40M wearable hours

Thu, May 1
- Paper2Code converts ML papers to code with 95% accuracy
- Eleven Labs launches AI Sound Effects in Studio editor
- Higgsfield adds Start & End Frames for controlled video storytelling
- VS Code now lets developers share UI elements with AI chats
- Claude Max subscription now includes Claude Code
- Neuralink gains FDA breakthrough status for speech-restoring BCI
- Ideogram 3.0 hits Freepik with designer-focused AI features
- FLORA now integrates Ideogram 3.0 for enhanced visual creation
- Ideogram 3.0 launches with realism boost and API access
- Cursor ranks Claude 3.7 Sonnet as top AI model for coding on April 2025
- Suno v4.5 launches with 8-minute songs and enhanced vocals
- ElevenLabs helps ALS patients maintain natural-sounding speech
- Pikaffects now lets users virtually try hairstyles instantly
- Anthropic launches Claude Integrations and enhanced Research mode
- Poe now integrates Gemini 2.5 Flash with 1M token support
- FutureHouse launches AI Scientist agents for superhuman research
- Krea launches GPT Paint for visual ChatGPT prompting
- Karpathy predicts visual AI interfaces will replace text chats
- Visa and Mastercard enable AI agents to make secure card purchases
- Genspark AI Slides adds rich editing with credit efficiency
- LM Studio now hosts Microsoft's new powerful Phi-4 reasoning models
- OpenAI retires GPT-4, preserves weights for historians
- Amazon Nova Premier hits Bedrock with 1M-token context
- Microsoft unveils powerful open-weight Phi-4 14B reasoning model
- Midjourney upgrades V7 model with new editor and aesthetics controls
- Meta's Llama Impact Grant expands AI access to offline communities
- NVIDIA CEO says China's AI capability trails US but remains competitive
- Google Gemini tests new AI image editing features
- Nvidia invests $500 billion in US AGI infrastructure development

Fri, May 2
- Google's Amplify Initiative tackles AI data gaps and creates health data platform worldwide
- Hedra Labs launches Character-3 API for developer builds
- Anthropic's Claude now offers smart web search worldwide
- Kling 2.0 now available on Poe for video creation
- Krea AI launches GPT Presets for action figures & 3D logos
- ElevenLabs launches EU data residency for enterprise AI audio
- Google releases 69-page prompt engineering guide for businesses
- Google brings ads to AI conversations amid revenue push
- LensAI video tool simplifies editing with adaptive AI features
- Google NotebookLM announces mobile app coming in weeks
- Midjourney's Omni-Reference lets users insert specific elements into images
- Google debuts multimodal AMIE for medical diagnostics

Sat, May 3
- Grok adds instant PDF generation for resumes and invoices
- Google's Gemini 2.5 Pro beats Pokémon Blue in AI gaming milestone
- NotebookLM now runs on Gemini 2.5 Flash for smarter answers
- Midjourney cuts V7 costs in half, boosts turbo speed