The 79
Posts
Gemini 2.5 Pro is here

Gemini 2.5 Pro is here

March 26, 2025

Hello pals! Here’s what you need to know about AI today:

👉 Google's most intelligent model, Gemini 2.5 Pro, is here

👉 OpenAI released a significant upgrade for ChatGPT’s image generation feature

👉 BMW and Alibaba are now AI partners

and many more!

📧 Did someone forward you this email? Subscribe here for free to get the latest AI news everyday!

Read time: 4.9 minutes

GOOGLE
Gemini 2.5 Pro is officially launched

Source: Google

What’s going on: Google launched Gemini 2.5, a new family of AI models designed to prioritize reasoning, marking a shift from traditional chatbot-style responses to more deliberate, problem-solving capabilities. The initial release, Gemini 2.5 Pro Experimental, is now available to users and developers via Google AI Studio and to Gemini Advanced subscribers for $20 a month.

What does it mean: According to google, Gemini 2.5 Pro is now officially the most intelligent Gemini model. This multimodal model handles text, images, audio, video, and code, aiming to outpace competitors like OpenAI’s o1 by pausing to “think” before responding, which boosts accuracy and performance on complex tasks. Google claims it excels in benchmarks like GPQA, AIME 2025, and SWE-Bench Verified, particularly in math, science, and coding, with a standout 68.6% score on the Aider Polyglot code-editing evaluation.

More details:

The model features a 1-million-token context window, expandable to 2 million soon, allowing it to process vast datasets, like entire codebases or lengthy documents, in one go.
On a test measuring software dev abilities, SWE-bench Verified, Gemini 2.5 Pro scored 63.8%, outperforming OpenAI’s o3-mini and DeepSeek-R1, but underperforming Anthropic’s Claude 3.7 Sonnet, which scored 70.3%.
On Humanity’s Last Exam, a multimodal test consisting of thousands of crowdsourced questions relating to mathematics, humanities, and the natural sciences, Google says Gemini 2.5 Pro scores 18.8%, performing better than most rival flagship models.
Interested and want to try it for free? Go to Google AI Studio and log in to your Google account → from the model dropdown select “Gemini 2.5 Pro Experimental 03-25“ → start using the model and enjoy!
Want to know about all the details and see the benchmarks? Read Google’s official blog post.

OPENAI
OpenAI introduces 4o image generation

Source: OpenAI | This image is AI-generated!

What’s going on: OpenAI has rolled out a significant upgrade to ChatGPT’s image-generation capabilities, integrating the GPT-4o model to enable native image creation and editing within the platform. Previously reliant on the separate DALL-E 3 model, ChatGPT now uses GPT-4o to produce more accurate and detailed images, leveraging the same multimodal system that handles text and code.

What does it mean: This shift allows for a unified approach where the GPT-4o can draw on its broad knowledge base to generate visuals that align closely with user prompts. The feature is already live for Pro subscribers at $200 a month, with a rollout planned soon for Plus, free users, and API developers.

More details:

The GPT-4o-powered image generation offers practical improvements over DALL-E 3, including sharper text rendering and the ability to handle complex prompts with up to 20 objects, far exceeding the typical 5-8 object limit of competing models.
Users can refine images iteratively through natural language instructions, such as adjusting backgrounds or transforming uploaded photos, making it useful for tasks like designing interfaces or creating educational diagrams.
OpenAI trained GPT-4o on a mix of public data and licensed content from partners like Shutterstock, aiming to balance capability with copyright considerations.
Interested in the details and want to see more examples? Read OpenAI’s official blog announcement.

🔍 Microsoft is adding Researcher and Analyst, two new AI-powered deep research tools to Microsoft 365 Copilot that can access both work data and the web to perform complex analyses.

🌏 Earth AI, a high-performance explorer company for clean energy minerals, has discovered promising deposits of critical minerals in previously ignored regions of Australia by utilizing AI.

💬 Quora's Poe chatbot app has launched two new subscription plans: a budget-friendly $5/month plan (10,000 points/day) and a high-end $250/month plan (12.5 million points) for heavy users. Poe allows access to multiple AI bots like DeepSeek-R1 and GPT-4o within a single platform.

🚗 BMW and Alibaba have expanded their partnership in China to enhance in-car AI using Alibaba's Qwen model, intensifying competition for Tesla, whose global sales are declining as local Chinese EV makers like BYD gain more popularity.

🤖 Google is expanding its AI-powered Search overviews to more European countries, with providing search experiences with concise summaries powered by generative AI.