🚀 DeepSeek Matches o3

PLUS: FLUX.1 Kontext AI Image Editor

Welcome back!

The open-source race just hit a new gear. The new DeepSeek R1-0528 matches Gemini 2.5 Pro and OpenAI o3 in logic, math, and code, all while staying open-source. With big jumps in reasoning, this release redefines what “free” AI is capable of. Let's unpack…

Today’s Summary:

  • 🔥 DeepSeek launches best open-source model

  • 👁️ FLUX.1 Kontext new state of the art image editor

  • 👓 Google debuts Android XR Glasses

  • 🗣️ Claude app gets Voice Mode

  • 🎓 Google tests Astra AI tutor

  • 🕹️ Spaitial trains 3D AI generation

  • 🛠️ 2 new tools

TOP STORY

DeepSeek becomes top open-source AI

The Summary: DeepSeek has released a new version of its open-source reasoning model, DeepSeek-R1-0528, bringing its performance to the level of Google Gemini 2.5 Pro and OpenAI o3. It shows remarkable improvements in logic, math, and code, while keeping the same original architecture. This release positions DeepSeek as the global leader in open-source AI.

Key details:

  • AIME 2025 test scores jumped from 70% to 87.5%, a 17.5-point leap in math reasoning

  • Coding benchmark LiveCodeBench rose from 63.5% to 73.3%, matching top-tier proprietary models

  • DeepSeek R1-0528 uses nearly double the reasoning tokens

  • Supports JSON output & function calling

  • Released under MIT License

  • A smaller distilled 8B version running on laptops is also available

Why it matters: DeepSeek is punching through a long-standing wall: free AI models are now matching the best from OpenAI and Google. R1-0528 proves that post-training, especially reinforcement learning, is now the fastest way forward, beating architectural changes and avoiding the cost of massive pretraining runs.

FLUX

Black Forest Labs launches FLUX.1 Kontext for image editing

The Summary: Black Forest Labs has released FLUX.1 Kontext, a new suite of image models that take both text and images as input for fast image editing. Unlike previous models, these models allow users to change styles, characters and scenes with precision, without distorting the original. Kontext works in real time and keeps edits coherent through iterations.

Key details:

  • FLUX.1 Kontext is up to 10x faster and cheaper than GPT-4o image generation, with near real-time output (~4 sec / image)

  • Retains pixel-level geometry, enabling precise edits like relighting and object swaps while preserving small details of the original

  • Character and style consistency is baked in; edits persist across multiple steps or entire scenes

  • Extremely good at colorizing old black-and-white photos

  • An open-source “dev” version will be released soon

  • Technical report

Why it matters: FLUX.1 Kontext cuts through the complexity of image editing pipelines and glue-code hacks like ControlNet and ComfyUI. It gives builders something they’ve craved: flexible, fast, high-fidelity AI image editing that maintains detail better than GPT-4o.

Test it for free in the FLUX Playground

FROM OUR PARTNERS

Start Your Day Smarter

Seeking impartial news? Meet 1440.

Every day, 3.5 million readers turn to 1440 for their factual news. We sift through 100+ sources to bring you a complete summary of politics, global events, business, and culture, all in a brief 5-minute email. Enjoy an impartial news experience.

GOOGLE

Google unveils Android XR Glasses

The Summary: Google unveiled Android XR-powered smart glasses that bring its Gemini AI assistant into everyday life. With a discreet in-lens display, voice-activated context recognition, and live translation, these glasses aim to be both wearable and useful. Built in collaboration with Warby Parker and Gentle Monster.

Key details:

  • Gemini AI now runs on Android XR glasses, offering voice-controlled navigation, messaging, photo capture, and translation

  • Google invested $150M in Warby Parker to co-develop the glasses, including a retail presence

  • Context-aware features include analyzing surroundings, remembering user tasks, and responding naturally

  • The single-lens display shows private information, making it possible to frame photos before you shoot, unlike Meta Ray-Bans

Why it matters: Smart glasses have mostly failed because they tried to do too much too early. Google’s new attempt learns from that. Offloading processing to the phone makes the hardware lighter and more wearable.

QUICK NEWS

Quick news

TOOLS

🥇 New tools

  • Tyce - Agent to craft deals, contracts, documents

  • Descript - AI video editor

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/