💡 OpenAI’s Latest Model Just Got Announced

Welcome, AI enthusiasts

OpenAI has introduced its powerful new o1-models (formerly 'Strawberry'), setting a new benchmark in reasoning and solving complex challenges in science, coding, and math. The capabilities of these models have been the center of attention in the AI community for months, and for good reason, as they represent a significant step forward in AI development. Meanwhile, Adobe is pushing creative boundaries with Firefly Video, expanding its generative AI to revolutionize video creation, while Mistral is making waves with Pixtral-12B, a multimodal model that handles both images and text. Let’s dive in!

In today’s insights:

• OpenAI’s Latest Model Thinks Before It Acts

• Adobe says video generation is coming to Firefly this year

• Mistral debuts Pixtral-12B multimodal AI model

Read time: 4 minutes

🗞️ LATEST DEVELOPMENTS

OPENAI
💡 OpenAI’s Latest Model Thinks Before It Acts

Evolving AI: OpenAI’s new o1-models (previously referred to as ‘Strawberry’) excel at reasoning and solving complex tasks in science, coding, and math. This first preview model, available now, reflects a major leap in AI capabilities.

Key Points:

• New reasoning models designed to think longer before responding.

• Significant improvement in solving complex tasks compared to previous models.

• Enhanced safety measures tested and implemented.

• o1-mini offers a faster, cost-effective solution for coding tasks.

Details:

OpenAI's new model, o1, marks a major step in AI scaling by increasing the compute time for inference, allowing the model to process questions more thoroughly before answering. This approach enhances reasoning capabilities, especially for logic-based tasks, by spending more time "thinking." While not universally superior to previous models like GPT-4o, o1 is designed for scenarios where deeper logical processing is valuable. OpenAI has also released o1-preview, a trial version, and o1-mini, a cost-effective model targeting STEM (Science, Technology, Engineering, and Mathematics) applications. These models outperform GPT-4 in science and math; for example, they scored 83% in the International Math Olympiad qualifier, while GPT-4 managed only 13%. Both are available now for ChatGPT Plus and Team users, with broader access planned soon.

Why It Matters:

This moment can’t be understated. Although a small beginning, this is a very important next step toward human-level intelligence. The idea is that if a model can go beyond pattern recognition and handle reasoning, it could unlock breakthroughs in areas like medicine, engineering, and possibly other challenges we haven’t been able to solve until now. The future is bright.

TOGETHER WITH FLOW
🗣️ Unlock 3x faster typing & AI prompts without lifting a finger (FREE)

Evolving AI: Tired of slow typing and endless edits? Flow (free) lets you speak naturally and converts your thoughts into perfectly formatted text, saving you hours.

→ Use your voice to craft ideal prompts in apps like ChatGPT, Cursor and v0

→ It works in every other application that you use.

→ With Flow’s advanced voice recognition, your tone and context are always captured

→ It eliminates mistakes and organizes your words seamlessly.

Boost your productivity today for free by using Flow →

ADOBE
🎥 Adobe says video generation is coming to Firefly this year

Evolving AI: Adobe is expanding its generative AI with Firefly Video, enhancing video creation and workflows.

Key Points:

• Adobe introduces text-to-video and image-to-video tools with simulated camera controls.

• Firefly AI model, initially available in beta, promises "commercial safety" from copyright issues.

• New features will be integrated into Adobe Creative Cloud apps later this year.

Details:

Adobe has revealed upcoming generative AI tools that create video clips from text descriptions or images. The tools, powered by the Firefly AI model, allow users to adjust angles, motion, and distance using simulated camera controls. The generated video quality is comparable to OpenAI’s Sora model, though limited to five seconds. Adobe emphasizes Firefly's commercial safety due to its use of licensed and public domain content. The beta release is planned for later this year, with future integration into Creative Cloud applications.

Why It Matters:

Adobe sees the Firefly Video Model as part of a broader push to integrate AI into creative workflows. The model is designed to handle various use cases, from generating 2D and 3D animations to creating atmospheric elements like smoke and fire. By incorporating generative AI into its suite of video editing tools, Adobe is positioning Firefly Video as an essential part of the modern editor’s toolkit—empowering creators to elevate their projects with the help of the latest in AI technology.

MISTRAL
🖼️ Mistral debuts Pixtral-12B multimodal AI model

Evolving AI: Mistral has introduced Pixtral-12B, a new open-source AI model capable of processing both images and text. This multimodal breakthrough is set to enhance image comprehension and text processing.

Key Points:

• Pixtral-12B, Mistral's first multimodal model, handles both images and text.

• It outperforms many open-source models but trails behind closed alternatives.

• The model is available for free under the Apache 2.0 license.

Details:

Pixtral-12B, boasting 12 billion parameters, can analyze images up to 1,024 x 1,024 pixels and decode Base64-encoded content. It supports OCR, diagram analysis, and satellite imagery processing, though video testing remains unconfirmed. Benchmarked against leading models, it performs well but falls short of closed models like GPT-4 in image comprehension. Available for free on GitHub and Hugging Face, it is released under Apache 2.0 for developers.

Why It Matters:

With the release of Pixtral 12B, Mistral will further democratize access to visual tools like content and data analysis. The model's exact performance is still uncertain, but this move continues the company's aggressive strategy in the AI field — and more competition is always better.

💡 Tip of the Day

We just had a real “wow” moment in AI.

Notebook LM by Google can generate engaging podcasts from your uploaded material for free. Within minutes, two agents will be discussing the content of your upload. The audio output is remarkably realistic and sounds convincingly human. It's hard to believe we’ve come this far already. We uploaded the text of our first topic on OpenAI's o1 — have a listen below.

Click here to try Notebook LM →

💰 OpenAI in talks to raise money at $150B valuation.

👀 Midjourney teases version 7 and external image editor.

🏷️ Facebook and Instagram are making AI labels less prominent.

📈 Trending AI Tools

• 🚀 LegalGraph - Empower teams to review contracts securely (link)

• 📽️ Hify - Create stunning sales videos from your browsesr (link)

• 💰 Expense Sorted - Use AI to categorize your expenses (link)

• 🎥 Kaiber - Turn text, videos, photos and music into videos (link)

• 🗣️ Deepgram - Build AI voices into your apps (link)