In partnership with

Welcome back!

Google isn’t slowing down as 2025 wraps up. Gemini 3 Flash, its new high-speed reasoning model, is 2x faster and 5x cheaper than GPT-5, all while taking the #2 spot on the global leaderboards. Let’s unpack…

Today’s Summary:

• 🚀 Google launches Gemini 3 Flash

• 🧾 Mistral unveils OCR 3 document AI

• 📈 2025 AI Year in Review

• 🧠 Anthropic open-sources Agent Skills

• 🌍 OpenAI launches ChatGPT Go low-cost tier

• 🧑‍💻 Developers can now submit ChatGPT apps

• 🛠️ 2 new tools

TOP STORY

Google launches Gemini 3 Flash high-speed reasoning model

The Summary: Google released Gemini 3 Flash, a new model focused on speed that has climbed to #2 on the global LMArena leaderboard just behind Gemini 3 Pro. It replaces 2.5 Flash across the Gemini ecosystem and also drives AI Mode in Google Search. The model stays within the Gemini 3 Pro intelligence tier while running at high speed and low cost.

Key details:

• Processes 218 tokens per second, 2x faster than GPT-5

• Now the default for the Gemini app and Google Search AI Mode

• Developers report it’s faster than Gemini 3 Pro with same quality

• Handles text, code, and video through full multimodal reasoning

Why it matters: Google is consolidating its lead not only with the smartest model but also with the smartest and fastest. Gemini 3 Flash begins to run in real time with the user queries, making high-level reasoning cheap enough to automate at scale. Google ended the year with a string of superior releases that tighten its lead on the AI frontier.

FROM OUR PARTNERS

Build Real AI Skills That Actually Pay Off

Build real AI and tech skills, faster

Udacity helps you build the AI and tech skills employers actually need—fast. Learn from industry experts through hands-on projects designed to mirror real-world work, not just theory.

Whether you’re advancing in your current role or preparing for what’s next, Udacity’s flexible, fully online courses let you learn on your schedule and apply new skills immediately. From AI and machine learning to data, programming, and cloud technologies, you’ll gain practical experience you can show, not just list on a résumé.

Build confidence, stay competitive, and move your career forward with AI and tech skills that are in demand.

Learn More

MISTRAL AI

Mistral launches OCR 3 AI

The Summary: Mistral AI has released OCR 3, a new document analysis model designed to extract text and structure from forms, handwriting, and complex tables with high precision. Mistral positions it to address the “paper bottleneck” holding back enterprise AI.

Key details:

• Under $1 per 1,000 pages for bulk jobs

• Can reconstruct tables with merged cells and multi-row headers

• Handles multilingual and handwritten documents

• Integrated with Mistral AI Playground and API

Why it matters: Mistral is betting that the real enterprise AI bottleneck is paper. The moat now is clean, structured data, and Mistral OCR 3 can turn static PDFs and handwritten content into precise structured data at scale. Companies still buried in scanned invoices and compliance forms shouldn’t remain locked out of the new data economy.

FROM OUR PARTNERS

The Fastest Way to Explain Anything

From Boring to Brilliant: Training Videos Made Simple

Say goodbye to dense, static documents. And say hello to captivating how-to videos for your team using Guidde.

1️⃣ Create in Minutes: Simplify complex tasks into step-by-step guides using AI.
2️⃣ Real-Time Updates: Keep training content fresh and accurate with instant revisions.
3️⃣ Global Accessibility: Share guides in any language effortlessly.

Make training more impactful and inclusive today.

The best part? The browser extension is 100% free.

Check out Guidde

YEAR IN REVIEW

2025 AI Year in Review

The Summary: Andrej Karpathy says that much of AI’s progress in 2025 came from Reinforcement Learning from Verifiable Rewards (RLVR), a new post-training technique used in "reasoning models". RLVR trains against problems with clear correct answers, like math puzzles, producing models that think through problems step-by-step.

Key details:

• Reinforcement Learning from Verifiable Rewards is a new technique introduced by OpenAI o1 and open-sourced by DeepSeek-R1, where models learn problem-solving through optimization and spontaneously develop reasoning steps

• Vibe Coding becomes real: The programming threshold crossed with Claude Code, Codex, and Cursor, where "code is suddenly free, ephemeral, malleable, discardable after single use".
  Y Combinator's Winter 2025 startup batch reported 25% of codebases were 95% AI-generated

• Adoption accelerates: 45% of US companies now use AI tools, up from 5% in 2023

• Multimodal breakthroughs: Google Gemini 3, Veo 3, and Nano Banana redefined multimodal capabilities

• Professional work redefined: A Yale study found no AI job losses despite widespread adoption. Work is evolving toward judgment, framing, and decision-making while speed and output are becoming commoditized

Why it matters: In 2025, AI moved from discovery to adoption. Developers started to realize the quality of reasoning AI tools is now production-ready, enterprises deploy for measurable outcomes, and knowledge workers explore how to use AI tools to augment themselves. The industry crossed from "can we build it?" to "how do we integrate it?".

QUICK NEWS

Quick news

• Anthropic open-sources Agent Skills as an industry standard

• OpenAI rolls out more affordable “ChatGPT Go” tier worldwide

• Developers can now submit apps to ChatGPT

TOOLS

🥇 New tools

• Google Disco - Turns open browser tabs into interactive mini-apps

• Claude in Chrome - Official extension to let Claude use your browser