Welcome, AI RESEARCHERS.

It's only been one week, and AI news has gotten crazy.

The developments from OpenAI, Anthropic, Figure, 1x Robotics, and Alibaba and much more. Here's everything that happened in AI over this weekend:

In today’s AI Researches:

• OpenAI Just Dropped GPT-4.5

• major AI developments this week

• ElevenLabs just launched Scribe

• Supercharge Your Research with Grok's DeepSearch

• AI Researches & AI pointers

• Trending Ai tools of the week

• More Ai news and developments of this week

Read time: 6 minutes

OPENAI

🚀 OpenAI Just Dropped GPT-4.5—Here’s What’s New

AI Researches: OpenAI has rolled out GPT-4.5 (codename Orion), a model focused on better understanding human emotions and intent—using unsupervised learning instead of traditional reasoning to boost its world knowledge.

The Details:

• GPT-4.5 offers a more fluid and natural conversation style, making interactions feel more intuitive and emotionally aware.

• While it’s more reliable and hallucinates less, it doesn’t outperform previous models in math or science, but it does surpass o3-mini and o1 in OpenAI’s freelance coding benchmark.

• The model is currently limited to Pro users and paid developers, with wider access rolling out next week.

• Pricing is a major shock: API access costs $75/$150 per million input/output tokens—far pricier than GPT-4o’s $2.50/$10 rate.

Why this Matters: GPT-4.5 feels like a personality refinement rather than a major technical leap. With steep pricing and modest improvements, it raises questions about OpenAI’s direction—are we seeing the peak of non-reasoning AI before a shift toward something entirely different?

Need a personal assistant? We do too, that’s why we use AI.

Ready to embrace a new era of task delegation?

HubSpot’s highly anticipated AI Task Delegation Playbook is your key to supercharging your productivity and saving precious time.

Learn how to integrate AI into your own processes, allowing you to optimize your time and resources, while maximizing your output with ease.

Get the free guide here.

Major Quick Developments

Alibaba’s Tongyi Lab introduced Wan2.1, an open-source suite of video generation models that excels in motion dynamics, physics simulation, and multi-modal tasks including text rendering in English and Chinese.

Anthropic released Claude 3.7 Sonnet, the first AI offering immediate responses with an extended thinking mode—displaying its reasoning via a scratchpad—and includes Claude Code, a command-line tool for editing, testing, and running code with precise token control.

Figure introduced Helix, an AI-driven approach that lets robots cooperate, respond to everyday spoken requests, and handle unfamiliar items with far less training than earlier methods.

Tencent has introduced Hunyuan Turbo S, a ‘fast-thinking’ AI model delivering instant replies at a lower cost, yet matching the performance of top competitors across knowledge, math, and reasoning tasks.

1X has launched NEO Gamma, next-gen home humanoid that demonstrates practical movement—including walking, squatting, and sitting—while using its in-house language model and ‘Emotive Ear Rings’ for clear communication, all within a safe, quiet design.

Alibaba's Qwen team introduced QwQ-Max-Preview, a reasoning-focused AI that reveals its thinking process during challenging tasks and will be available as open-source under Apache 2.0.

ELEVENLABS

⚡ ElevenLabs just launched Scribe, an AI that transcribes speech better than ever

AI Researches: ElevenLabs just dropped Scribe, a speech-to-text AI that claims to be the most accurate in the world, outperforming OpenAI’s Whisper v3 and Google’s Gemini 2.0 Flash in multiple languages.

The Details:

• Scribe understands 99 languages and boasts over 95% accuracy in 25+ of them, including English, Spanish, and Italian.

• It excels in transcribing underrepresented languages like Serbian, Cantonese, and Malayalam, where speech recognition is often unreliable.

• Features include multi-speaker labeling, word-level timestamps, and detection of non-verbal sounds like laughter and music.

• Pricing starts at $0.40 per hour for pre-recorded audio, with a real-time version coming soon.

AI TUTORIAL

✅ Supercharge Your Research with Grok's DeepSearch

AI Researches: Use Grok’s DeepSearch to quickly scan multiple websites and find the latest studies in just a few minutes.

Step-by-Step Guide:

• STEP 1: Access DeepSearch

  • Log in to your Grok account (on web or via X).

  • Look for the “DeepSearch” button in the main interface and click it.

• STEP 2: Enter Your Query

  • Type a straightforward request, for example: “Summarize the latest findings in [topic].”

  • This tells DeepSearch exactly what to look for across many websites.

• STEP 3: Review the Results

  • Wait a short moment for DeepSearch to gather information.

  • You’ll get a clear report showing major discoveries, key players, and direct links to more details.

• STEP 4: Ask Follow-Up Questions

  • If something grabs your attention, enter a new prompt to learn more (e.g., “Explain the methods used in [specific discovery].”).

  • This targeted approach helps you dig deeper and get exactly the details you need.

AI Research Roundup

Microsoft Research has released BioEmu-1, an AI system that generates thousands of protein structure samples per hour, reducing months of computation to minutes and offering high prediction accuracy

Google's Co-Scientist AI matched a decade-long study on bacterial resistance in just 48 hours by independently confirming key experimental findings.

In a reported advancement from Fudan University, Chinese scientists have introduced an AI-enabled intervention for Parkinson’s disease, signaling a notable shift toward faster-paced medical discoveries following decades of incremental progress.

AI Pointers

Sam Altman believes that AI agents capable of delivering $5,000 worth of work for just 50 cents of compute represent an immense economic opportunity, describing it as “lightning in a bottle.”

Mark Cuban suggests that if he were 12 again, he would spend time reading and mastering AI, as these skills not only build a strong knowledge base but also prepare one for a future where job requirements are set to evolve.

• 🔧 Gemini Code Assist – A free AI coding assistant with 180K code completions per month

• 🎞️ Project Starlight by Topaz Labs – Restore and enhance old videos using AI

• 🪵 Granite 3.2 – IBM’s lightweight open models tailored for enterprise solutions

• 🎙️ Octave TTS – Create AI-generated voices with expressive emotional delivery

• ⚡ Zapier Agents – Automate workflows across 7,000+ apps using AI-driven agents with internal data

• 📺 SmolVLM2 – Compact AI models for video analysis on mobile devices and laptops

• 🎧 ElevenLabs Studio – Edit, structure, and generate high-quality long-form audio

Pika Labs has launched its updated 2.2 model, boasting enhanced quality, 10-second 1080p resolution generations, and new capabilities for transitions and transformations.

Microsoft has introduced new updates to Copilot, featuring a dedicated MacOS app, support for uploading PDF and text files, and several UI enhancements.

Meta unveiled Aria Gen 2 glasses equipped with advanced sensors, on-device AI processing, and an all-day battery designed for research in machine perception, contextual AI, and robotics.

You Labs launched ARI, a research agent capable of analyzing up to 400 sources and generating professional reports—with charts, citations, and visuals—in under five minutes.

Ideogram has introduced its updated 2a text-to-image model, which produces images in just 10 seconds—with a Turbo option at twice the speed—while lowering costs by half and enhancing graphic design and photorealism.

DeepSeek reportedly generates around $205 million in annual revenue, with daily revenue of $562,000 and costs of $8,000, suggesting a substantial profit margin.

Two developers have created Gibber Link, a protocol that uses sound signals for AI agents to identify each other during calls and switch from human speech to data transmission during calls, resulting in up to 90% lower compute costs and 80% faster communication.

Pika Labs has launched Pika Swaps, a tool that lets users replace any element or character in a scene using image or text prompts.

Hume AI has unveiled Octave, a text-to-speech LLM that grasps emotional context, allowing creators to design custom voices with precise control over emotion and delivery.

Poe has introduced Poe Apps, a platform that lets users build applications and visual interfaces by harnessing a blend of reasoning, multimodal, image, video, and audio models.

Vevo Therapeutics has unveiled the Arc Virtual Cell Atlas, featuring Tahoe-100M—an open-source dataset that maps 60,000 drug-cell interactions across 100 million cells.

Exa launched Websets, a search product that employs advanced agents to deliver superior results, outperforming Google by over 20 times and OpenAI Deep Research by 10 times on complex queries.

OpenAI has introduced its new Deep Research feature for ChatGPT Plus, Team, Edu, and Enterprise tiers, offering 10 queries per month—while the $200 Pro tier receives 120 queries.

Anthropic is reportedly preparing for a larger-than-anticipated $3.5B funding round at a $61.5B valuation, with this news emerging just days after the release of Sonnet 3.7.

DeepSeek, meanwhile, is reportedly planning to accelerate the launch of its next AI model—moving up its originally scheduled May release to build on the success of R1.

Additionally, OpenAI unveiled an Advanced Voice feature powered by GPT-4o mini, making it available for free to ChatGPT users while promising the same conversational style offered in the GPT-4o version for Plus and Pro subscribers.

Share Your Opinion