🇨🇳 China's MiniMax crushes Opus 4.6

Good Morning! MiniMax just dropped M2.5, an open-source coding model that’s punching at frontier weight, without the frontier price tag. Plus, I’ll show you how to turn any ordinary photo into a cinematic, photorealistic masterpiece.

Plus, in today’s AI newsletter:

• MiniMax Releases M2.5 Coding Model

• OpenAI Launches Codex-Spark With Custom Hardware

• Gemini 3 Deep Think Breaks Reasoning Benchmarks

• How to Turn Any Ordinary Photo Into a Cinematic Masterpiece

• 4 new AI tools worth trying

AI MODELS

⚙️ MiniMax Releases M2.5 Coding Model

MiniMax unveiled M2.5, a 100% open-source model with just 10B active parameters that rivals closed giants like Opus and GPT-5.x, at a fraction of the cost.

• Competitive with Opus 4.6 / GPT-5.2 in coding, agents, search, and office work

• Runs at up to 100 tokens/sec, ~3× faster thinking efficiency than Opus

• $0.3/M input, $1.2/M output, up to 20× cheaper than frontier models

• SOTA results on SWE-Bench, BrowseComp, and real-world agentic tasks

M2.5 shows that frontier-level performance no longer requires closed models or massive costs. This could be another “DeepSeek moment,” accelerating open-source dominance and making powerful AI agents affordable at scale.

AI MODELS

⚡ OpenAI Launches Codex-Spark With Custom Hardware

OpenAI introduced GPT-5.3-Codex-Spark, a lightweight version of Codex built for ultra-low latency coding and rapid iteration, backed by custom hardware from Cerebras.

• Codex-Spark is optimized for real-time collaboration and fast prototyping

• Runs on Cerebras’ Wafer Scale Engine 3, a megachip with 4 trillion transistors

• Available in research preview for ChatGPT Pro users via the Codex app

• Marks the first major milestone in OpenAI’s $10B+ multi-year Cerebras partnership

This is a clear signal that OpenAI is vertically integrating, not just building models, but pairing them with purpose-built chips. Faster inference unlocks new coding workflows, real-time agent collaboration, and a fundamentally different developer experience.

AI MODELS

🧠 Gemini 3 Deep Think Breaks Reasoning Benchmarks

Google’s specialized reasoning model, Gemini 3 Deep Think, scored 84.6% on ARC-AGI-2, beating GPT-5.2 and Claude by a wide gap. ARC-AGI-2 is designed to resist memorization and test real, on-the-fly reasoning.

• Gemini 3 Deep Think: 84.6%, GPT-5.2: 52.9%, Claude Opus 4.6: 68.8% (humans ~60%)

• Also hit 48.4% on Humanity’s Last Exam and 3,455 Elo on Codeforces (elite human tier)

• Achieved gold-medal-level results in 2025 math, physics, and chemistry Olympiads

• Verified inference cost is $13.62 per task, raising scale and economics questions

This isn’t a small win, it’s a clear leap in reasoning ability. Gemini 3 Deep Think shows that with more test-time compute, AI can outperform humans on tasks built to block shortcuts. The tradeoff now shifts from capability to cost.

HOW TO AI

🗂️ How to Turn Any Ordinary Photo Into a Cinematic, Photorealistic Masterpiece

In this tutorial, you’ll learn how to transform a simple, everyday photo into a high-end, cinematic, editorial-style image using a single prompt inside Google Gemini. No expensive camera gear, no Photoshop skills, and no complex workflows, just one well-crafted prompt that does the heavy lifting for you.

🧰 Who is This For

• Creators who want luxury, cinematic visuals without pro equipment

• Photographers looking to enhance images instantly

• Designers and marketers who need premium-looking visuals fast

• Anyone who wants photorealistic results with minimal effort

STEP 1: Pick Your "Base" Photo

Before we touch the AI, pick the right starting point. Because our prompt asks for "Riviera chic" (think vintage summer, luxury, and bold colors), the best photos to use are:

• Outdoor shots with natural light.

• Photos where the subject’s face is clear.

• Vacation or outfit photos.

STEP 2: Upload to Google Gemini

Open gemini.google.com or open the Gemini app.

1. Look for the Image icon in the chat bar.

2. Select your chosen photo from your device.

3. Once the thumbnail appears in the chat box, you are ready for the magic.

STEP 3: Deploy the "Masterpiece" Prompt

Now, copy and paste the specific prompt below into the message box with your image. This prompt is engineered to keep your face and clothes 100% accurate while changing the vibe and quality.

Copy this exactly:

"Transform the source image by applying a vibrant, retro 1960s 'Riviera chic' aesthetic, characterized by a striking color contrast between deep teals and bold reds. The lighting should mimic bright, harsh mid-day sunlight, creating strong, defined shadows and high saturation. The overall atmosphere should feel like a vintage high-fashion magazine editorial or a classic summer film still, precisely maintaining the original subject's facial features, pose, outfits, background, and gender 100% accurate. Perform a realistic upscale, bringing the subject into extremely sharp focus with enhanced skin texture and detail."

STEP 4: Refine and Download

Hit Enter and wait a few seconds. Gemini will process the image and generate a new version.

• Check the Details: Look at the skin texture and the "Teal and Red" contrast. It should look significantly more "expensive" than the original.

• The "Upscale" Effect: Notice how the focus is sharper.

• Final Step: Click the Download icon on the image you like best.

Why this works: > Most AI image generators change your face too much. This specific prompt uses "100% accurate" constraints, which forces Gemini to act more like a high-end filter and professional retoucher rather than just making a random person.

Anthropic raised a $30B Series G led by GIC and Coatue and co-led by D. E. Shaw, Dragoneer, Founders Fund, Iconiq, and MGX at a $380B post-money valuation.

ByteDance's new AI video generation model Seedance 2.0 goes viral in China, with one state-backed newspaper saying it is bigger than DeepSeek's “Sputnik moment”.

In a memo to US lawmakers, OpenAI accused DeepSeek of using distillation techniques to train the next generation of R1 and “free-ride” on leading US AI models.

Global installed base of active smartphones grew 2% in 2025; eight OEMs had 200M+ active devices each, and nearly one in four active smartphones is an iPhone.

🧑‍💻 GLM-5: Ziphu AI’s new open-source frontier model

🧑‍💻 Claude Opus 4.6: Built for long, serious work in coding and research

⚙️ GPT-5.3 Codex: Faster coding model with strong reasoning

🎨 Qwen-Image-2.0: Alibaba’s all-in-one image generation and editing model

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas