o1-pro API: Most expensive OpenAI API yet!

Along with - NVIDIA GTX Updates

Hi there 👋

Nvidia GTC 2025 is shaping up to be the Apple WWDC of AI. 

Jensen Huang, in his signature black leather jacket, took the stage to unveil robots, self-driving cars, next-level CGI, and Nvidia’s latest Blackwell system.

These updates come at a time when Nvidia is under pressure. Back in January, its stock plummeted 17%, wiping out nearly $600 billion in market value- the biggest single-day loss for any U.S. company in history.

The trigger was DeepSeek. The Chinese AI model that delivered high performance at a fraction of the cost, shaking up the AI hardware landscape.

But Jensen seems to be betting big on enterprise AI demand, doubling down on Nvidia’s future. And if GTC is any sign, the AI arms race is only accelerating.

What would be the format? Every week, we will break the newsletter into the following sections:

  • The Input - All about recent developments in AI

  • The Tools - Interesting finds and launches

  • The Algorithm - Resources for learning

  • The Output - Our reflection

Table of Contents

NVIDIA’s GTC 2025 is underway, with major announcements spanning AI, robotics, autonomous vehicles, and reasoning models. Here’s a rundown of the latest updates:

1. Blackwell Architecture - The Next Leap in AI Computing

  • Nvidia introduced Blackwell GPUs, delivering 1 exaflop of computing power in a single rack.

  • The roadmap includes:

    • Blackwell Ultra NVL72 (H2 2025).

    • Vera Rubin NVL144 (H2 2026) - named after the scientist who discovered dark matter.

    • Rubin Ultra NVL576 (H2 2027) - pushing 600kW per rack.

2. AI Factories - The Future of Data Centers

  • Huang highlighted the shift from traditional data centers to AI factories- high-performance setups optimized for industrial-scale AI.

  • Networking innovations like Spectrum-X Ethernet and silicon photonics (1.6 Tbps bandwidth) aim to solve scaling challenges.

3. Nvidia Dynamo - The AI OS for Large-Scale Inference

  • Nvidia announced Dynamo, an AI operating system designed to help enterprises deploy full-stack AI solutions at scale.

4. Robotics & Digital Twins

  • Nvidia partnered with Google DeepMind and Disney to introduce Newton, a physics engine for real-time robotic training and digital twin simulations.

5. Enterprise AI - The Rise of AI Agents

  • Huang predicted that AI agents will play a critical role in business operations.

  • Forecast: 10 billion AI agents globally by year-end, with Nvidia's own operations being fully AI-assisted.

6. Automotive Innovations

  • Nvidia’s Halos system will enhance self-driving safety. General Motors is integrating Nvidia tech into its autonomous vehicle fleet.

What This Means for the Future

Huang positioned Nvidia as the driving force behind AI’s evolution- not just as an application, but as the foundation of computing itself. From AI factories to enterprise-scale automation, Nvidia is doubling down on scalability, efficiency, and AI-powered transformation across industries.  (source)

OpenAI has launched o1-pro, a more powerful version of its o1 reasoning model, now available via API. The model uses more compute to deliver better problem-solving performance, but comes at a steep cost:

  • $150 per million input tokens (~750,000 words)

  • $600 per million output tokens (10x the price of o1)

Currently limited to developers who’ve spent at least $5 on OpenAI API services, o1-pro aims to provide more reliable answers to complex problems. However, early user feedback has been mixed, with marginal improvements over o1 in coding and math, and some struggles with logic puzzles like Sudoku. (source)

A few days ago, Google DeepMind introduced Gemma 3, and we  were still exploring its capabilities. But now, there’s a major development - Mistral AI has dropped Small 3.1, and it’s making bold claims. This lightweight, fast, and highly customizable model is designed to deliver top-tier performance while running effortlessly on a single RTX 4090 or a Mac with 32GB RAM, making it ideal for on-device applications.

What’s New in Mistral Small 3.1?

  • Compact but Powerful - Despite being a small model, it competes with much larger models like Gemma 3 and GPT-4o Mini, offering faster inference (150 tokens/sec) and a 128K context window.

  • Multimodal Capabilities - It can handle text and images, making it a versatile tool for developers.

  • Low-Latency Function Calling - Ideal for automation and AI-driven workflows, ensuring faster and more responsive interactions.

  • Fine-Tuning Ready - Can be customized for legal, medical, security, and other domain-specific applications.

The model is open-source under the Apache 2.0 license and is available on Hugging Face, Mistral AI’s developer platform, and Google Cloud Vertex AI. It will soon be accessible on NVIDIA NIM and Microsoft Azure AI Foundry as well.

Mistral is clearly positioning Small 3.1 as the best model in its weight class, offering a balance of efficiency, flexibility, and speed. If you’re looking for a powerful AI model that doesn’t require heavy compute, this might be worth checking out. (source)

Baidu has just dropped two major AI models- ERNIE 4.5 and X1, making some bold claims. They’re positioning these models as more powerful than OpenAI’s GPT-4.5 and even cheaper than DeepSeek-R1. If the benchmarks hold up, this could be a game-changer in AI affordability and performance.

ERNIE 4.5: Multimodal & Emotionally Intelligent

  • A true multimodal model - Handles text, images, audio, and video, making it versatile for different applications.

  • Stronger language capabilities - Designed for more natural conversations and long-term memory retention.

  • High emotional intelligence - Can interpret memes and nuanced text, a step toward better contextual understanding.

  • Performance claims - Baidu says ERNIE 4.5 beats GPT-4.5 across multiple benchmarks while being significantly more cost-efficient.

ERNIE X1: The Deep-Reasoning Model

  • Optimized for complex problem-solving - Built for logical reasoning, planning, and structured thought.

  • Designed to mimic human-like reasoning - More than just a language model, it focuses on how information is processed and applied.

  • Performance vs. Cost - Baidu claims X1 matches DeepSeek-R1 in performance but at half the cost.

Baidu is making a strong push to reclaim its dominance in China’s AI race, where startups like DeepSeek have been gaining ground. Following the announcement, Baidu’s stock surged by 10%, signaling strong market confidence in these models. (source)

OpenAI just gave its o1 and o3-mini models a serious boost- they can now run Python-powered data analysis directly inside ChatGPT. This means you can perform statistical regressions, visualize business metrics, and even run simulations- all within a chat.

No need for external tools- just drop in your data and let ChatGPT handle the heavy lifting. Whether it’s financial forecasting, A/B testing, or exploratory analysis, this update makes advanced data insights way more accessible.

Google is transitioning Google Assistant users on mobile to Gemini, integrating AI-powered capabilities. Over the next few months, Gemini will replace Assistant on most mobile devices, tablets, cars, headphones, and watches. A Gemini-powered experience for home devices like speakers, displays, and TVs is also in development.

New Features in Gemini

  • Canvas - A tool for writing and editing documents and code within Gemini, with real-time collaboration and Google Docs export.

  • Audio Overview - Converts documents and slides into podcast-style discussions, summarizing content dynamically.

  • Enhanced AI Capabilities - Includes Gemini Live for multimodal conversations and Deep Research for AI-driven information gathering. (source) (source)

At Adobe Summit 2025, Adobe rolled out a set of AI-driven innovations aimed at revolutionizing customer experience orchestration (CXO)- blending AI seamlessly into marketing, content creation, and customer engagement.

Key Announcements

  • Experience Platform Agent Orchestrator - Manages AI agents across Adobe and third-party platforms to optimize automation and interactions.

  • Generative AI Expansion - Firefly AI models are now embedded into Adobe’s content workflows, with GenStudio streamlining content production.

  • Enterprise AI Agents - New AI-powered tools assist with content creation, data management, website optimization, and B2B engagement.

  • Content Analytics - Real-time tracking to measure and optimize content performance for better engagement and conversions.

  • AI for Video & 3D - Firefly Services APIs enable automated translation, lip-syncing, video resizing, and 3D asset generation for marketing and e-commerce.(source)

Figure Unveils BotQ: A High-Volume Humanoid Robot Manufacturing Facility

Figure has launched BotQ, a new facility designed for large-scale humanoid robot production. The first-generation manufacturing line is capable of producing up to 12,000 robots per year, with plans for significant expansion.

Key Developments in BotQ

  • Vertically Integrated Manufacturing - Figure is bringing production in-house to maintain quality and efficiency.

  • Scalable Robot Design - The new Figure 03 model is optimized for mass production, shifting from CNC machining to injection molding and diecasting for faster, cost-effective manufacturing.

  • AI & Automation in Manufacturing - Figure’s humanoid robots will assist in their own production, handling material transport and assembly tasks using internal AI systems.

  • Advanced Reliability Testing - BotQ includes a dedicated reliability team using accelerated lifecycle tests to improve robot durability.

  • Supply Chain Development - Figure is building its own supply chain, designing core components like actuators, sensors, batteries, and electronics while partnering with vendors for specialized parts.

  • Custom Manufacturing Software - The facility runs on a proprietary Manufacturing Execution System (MES), integrating real-time tracking, quality control, and IoT monitoring.

The Future of BotQ

With plans to scale up to 100,000 robots, BotQ represents a major shift in humanoid robot production, integrating automation and AI to streamline manufacturing. (source)

Roblox has launched Cube, an AI-powered 3D model generation tool, allowing creators to generate 3D meshes from simple text prompts. The model is now open source, enabling developers outside the platform to customize and expand its capabilities.

Key Features & Future Plans

  • Mesh Generation (Beta) - Converts text prompts into 3D objects, editable within Roblox Studio.

  • Upcoming AI Tools - Text generation, text-to-speech, and speech-to-text will be introduced in the coming months.

  • Future Enhancements - Plans include complex object and scene generation, eventually leading to “4D creation”, where AI-generated objects interact dynamically. (source)

  • LG AI Research has unveiled EXAONE Deep, a high-performing AI excelling in maths, science, and coding, rivaling top-tier models despite its smaller size. Its 32B model matched DeepSeek-R1 (671B) in the AIME 2025 exam and led in global math benchmarks, while also outperforming in doctoral-level physics, chemistry, and biology (GPQA Diamond) and scoring 59.5 on LiveCodeBench for coding. Additionally, it achieved 83.0 on the MMLU benchmark, ranking as Korea’s top domestic model. Recognized as a "Notable AI Model" by Epoch AI, LG remains the only Korean entity on the list in two years. (source)

  • Boston Dynamics has unveiled the latest progress in the movement capabilities of its humanoid robot, Atlas. Known for its agility and balance, Atlas has demonstrated enhanced mobility, pushing the boundaries of dynamic locomotion and real-world adaptability. While details on the exact improvements remain limited, Boston Dynamics' ongoing developments suggest greater precision, stability, and dexterity, reinforcing Atlas’ role in robotics research and industrial applications.

Flowsage: AI-Powered Flowchart Creation

Flowsage revolutionizes flowchart creation, making it easier than ever to visualize ideas, map out processes, and navigate complex workflows.

How to Use Flowsage

  1. Sign Up - Create an account on the Flowsage website.

  2. Choose a Plan - Select either a free or paid subscription based on your needs.

  3. Write a Prompt - Describe the flowchart you need, and Flowsage will generate it instantly.

  4. Edit & Customize - Make manual adjustments and modifications as required.

  5. Share & Optimize - Collaborate with others and refine your workflow efficiently.

Last week, we launched another free AI & ML course to help you gain hands-on experience with cutting-edge AI tools and frameworks-

  • Demystifying OpenAI Agents SDK- Learn to navigate and implement OpenAI's Agents SDK to build autonomous AI workflows. This course explores the core functionalities of OpenAI Agents SDK, including agent orchestration, tool integration, and real-world applications. Gain hands-on experience in developing AI-driven assistants that execute tasks, retrieve information, and interact dynamically within various environments.

  • Deep Dive Into QwQ-32B- Master the QwQ-32B AI model, a groundbreaking advancement in deep learning. This course unpacks its architecture, implementation, and practical applications, empowering professionals to leverage its full capabilities for real-world AI solutions.

This week GTC 2025 had some incredible announcements, but the AI hardware race is just getting started. With a lot of Chinese players it’ll be interesting to see how Nvidia and other American counterparts adapt.

Until next time.

Reply

or to participate.