In this edition we’ll be covering…
Z.AI's GLM-4.7 open-source model taking on OpenAI, Anthropic, and Google in coding benchmarks
How to build professional landing pages using Claude Opus 4.5's artifact capabilities
Instacart's abrupt end to AI pricing experiments after charging customers up to 23% more for the same items
5 breaking AI developments from hiring chaos to Yann LeCun's new startup
3 productivity tools from logo animation to voice-matching writing assistants
(BONUS) Expert Christmas image prompts that will blow your friends and family away
And much more…
The Latest in AI
China’s Open-Source Coding Challenger Takes on the Big Three
China's Z.AI just threw its hat in the ring with GLM-4.7, an open-source model that's turning heads in the coding world. While everyone's been watching OpenAI, Anthropic, and Google duke it out, Z.AI quietly built something that scored 42% on the Human-Like Evaluation (HLE) benchmark—a 38% jump from their previous GLM-4.6.
The timing couldn't be more interesting. As Western AI labs keep their best models locked behind API walls, China's making a power move with open-source...
Key Performance Highlights:
LiveCodeBench V6: GLM-4.7 scored 84.8, surpassing Claude 4.5 Sonnet in coding tasks—a notable achievement for an open-source model competing against proprietary systems.
LMArena Code Arena: Clinched the number one spot among open-source models, outperforming GPT-5.2 in blind tests where developers rated code quality without knowing which model generated it.
Agentic Coding: Achieved a 90.6% tool-calling success rate, beating Claude-4-Sonnet (89.5%), Kimi-K2 (86.2%), and Qwen3-Coder (77.1%) when integrated with coding frameworks across 52 different tasks.
So What?
This isn't just another model launch, it's a statement about where AI development is headed. While Western companies chase AGI behind closed doors, China's betting on an open playbook. GLM-4.7 proves you don't need a $10 billion training run to compete at the frontier of coding AI.
The open-source release means developers can run this locally, fine-tune it for specific use cases, and integrate it into their workflows without API costs stacking up. For teams building coding agents or automated development tools, this levels the playing field in ways we haven't seen since the early days of GPT-3.
If a relatively unknown Chinese lab can match Claude and GPT on coding benchmarks with an open model, what does that say about the supposed "moat" of the AI giants? The race just got a whole lot more interesting…
Get Your Hands Dirty!
Building Landing Pages with Claude Opus 4.5
Claude Opus 4.5 is specifically designed for complex, multi-step work that requires both reasoning and polish—making it perfect for creating production-ready landing pages that actually look professional.
Here's how to leverage Opus 4.5's strengths to create a landing page artifact that's ready to ship:
Step 1: Select Claude Opus 4.5
Open Claude.ai and switch to the Opus 4.5 model from the model selector. You'll find it next to Sonnet 4.5 and Haiku 4.5. Remember: Opus prioritizes depth and quality over speed, which is exactly what you want for design work.
Step 2: Use This Detailed Prompt
Copy and paste this prompt structure, customizing it for your needs:
Create a modern, conversion-optimized landing page for [PRODUCT/SERVICE NAME] - a [ONE-SENTENCE DESCRIPTION].
Target Audience: [WHO IS THIS FOR]
Main Goal: [PRIMARY ACTION YOU WANT VISITORS TO TAKE]
Design Requirements:
- Color palette: [SPECIFY COLORS OR SAY "modern tech brand" or "warm and approachable"]
- Typography: Clean, readable, professional
- Layout: Single-page with smooth scroll sections
- Responsive: Must work perfectly on mobile
Sections to Include:
1. Hero Section: Compelling headline, subheadline, and clear CTA button
2. Problem Statement: What pain point does this solve?
3. Solution/Features: 3-4 key features with icons
4. Social Proof: Testimonial section (use placeholder quotes)
5. Pricing/CTA: Final conversion section
6. Footer: Links and contact info
Style Preferences:
- [modern/minimal/bold/playful]
- Generous whitespace
- Smooth animations on scroll
- Professional imagery placeholders
Please create this as a complete HTML artifact with embedded CSS and JavaScript. Make it polished and production-ready.Step 3: Review and Iterate
Opus 4.5 will generate a complete, working landing page. Here's where its strengths shine:
Visual components come out more considered than previous models. The spacing, hierarchy, and overall feel need less refinement.
Everything should work immediately—no broken layouts or missing styles.
Test it on mobile by resizing your browser. Opus 4.5 builds responsive behavior in from the start.
Step 4: Refine Specifics
If something's not quite right, give targeted feedback:
"Tighten the spacing in the hero section"
"Make the CTA button more prominent with a shadow effect"
"Add a subtle animation when features scroll into view"
"Can you adjust the color palette to be more vibrant?"
Opus 4.5 is particularly good at understanding loose instructions and figuring out the intent behind your feedback.
🔥 Power Tip: Upload a screenshot of a landing page you admire and ask Opus to "match this aesthetic" instead of describing the style you want. This is significantly more effective than trying to articulate design preferences in words.
This is just the beginning of leveraging Claude to its fullest capabilities. If you want to truly become a Claude master, check out the module on our platform!
Industry Intel
Instacart Pulls the Plug on AI Price Experiments
Instacart just learned a hard lesson about algorithmic pricing: customers really don't like finding out they're paying more than the person next to them for the exact same eggs.
After a Consumer Reports investigation revealed that Instacart was charging different users different prices for identical items—sometimes up to 23% more—the company abruptly ended its AI-driven pricing experiments. One particularly egregious example? The same brand of eggs at the same store was priced at five different points: $3.99, $4.28, $4.59, $4.69, and $4.79.
The company insisted these were just "tests" to understand shopper behavior, but the timing tells a different story...
What Went Wrong:
A CR survey found that 72% of respondents considered differential pricing unacceptable, regardless of the reason. When shoppers discovered they were paying more than others, trust evaporated.
Representative Robert Garcia demanded documents and information about Instacart's use of algorithmic pricing by January 2, 2026, citing concerns about "surveillance pricing" using personal and demographic data.
Instacart has filed patent applications mentioning the use of personal, behavioral, and demographic data to determine prices—raising red flags about whether certain users were systematically charged more based on their profiles.
So What?
This is a watershed moment for AI in retail pricing. While airlines and hotels have used dynamic pricing for years, groceries hit different. When you're charging working families different amounts for milk and eggs based on an algorithm's assessment of what they'll tolerate, you've crossed a line.
The episode also highlights a critical tension in AI deployment: just because you can personalize pricing doesn't mean you should. The technology worked exactly as designed—Instacart successfully identified price sensitivity and adjusted accordingly. But the business and ethical implications weren't fully considered.
Quick Bites
Stay updated with our favorite highlights, dive in for a full flavor of the coverage!
Meta's Yann LeCun is targeting a €3 billion ($3.5 billion) valuation for his new startup Advanced Machine Intelligence (AMI) Labs, which aims to create "world models" that understand physics and can plan complex actions.
AI hiring has created a "doom loop" where both companies and job seekers are miserable—applicants use ChatGPT to apply to hundreds of jobs while companies deploy AI to auto-reject them, resulting in lower hiring rates and starting wages according to Dartmouth research.
YouTube launched Playables Builder, an AI-powered tool using Gemini 3 that lets creators make bite-sized games from short text, image, or video prompts, currently in closed beta for US, Canada, UK, and Australia.
OpenAI admits that AI browsers may always be vulnerable to prompt injection attacks—describing it as "unlikely to ever be fully solved," much like scams and social engineering, even as they work to harden ChatGPT Atlas.
Developers can now submit apps to ChatGPT for review and publication in the new app directory at chatgpt.com/apps, extending conversations with new context and actions like ordering groceries or creating slide decks.
Trending Tools
🎬 Logo Animation - Transform static logos into stunning motion graphics in seconds with AI-powered animation styles.
✍️ Tonemark - The AI writing assistant that learns your unique writing style from your best posts and generates LinkedIn and X content that actually sounds like you, with built-in citation to your uploaded docs.
🛠️ ChatGPT Apps SDK - Build chat-native experiences that bring context and action directly into ChatGPT conversations, now accepting submissions for the official app directory with monetization options coming soon.
The Neural Network
Tired of generic Christmas cards? We thought it would be fitting to share our best prompts to generate amazing, festive images with your favorite image models.
Here's how to get images that don't look like they came from a stock photo database…
Cozy Christmas Morning
A warmly lit living room on Christmas morning, golden hour sunlight streaming through frost-covered windows, illuminating a beautifully decorated tree with handmade ornaments and soft white lights. A golden retriever puppy sleeps peacefully on a plush rug near the fireplace where stockings hang. Wrapped presents with vintage-style paper and velvet ribbons are scattered beneath the tree. Steam rises from two mugs of hot chocolate on a wooden coffee table. Shot in the style of a Kinfolk magazine editorial, soft focus, warm color grading, cozy and intimate atmosphere.Winter Wonderland Portrait
A candid portrait of a family laughing together while building a snowman in their backyard at dusk, string lights from the house creating bokeh in the background. Everyone wears chunky knit sweaters in cream and burgundy tones. Gentle snow is falling, captured mid-air with a shallow depth of field. The scene has the warm, nostalgic quality of a film photograph from a Nikon F3, slightly grainy, natural colors with a slight magenta shift in the shadows, golden hour lighting despite the snow.Elegant Holiday Dinner
An overhead shot of an elegantly set dinner table for the holidays, featuring a centerpiece of fresh pine branches, white candles of varying heights, and scattered pomegranates. Fine china with gold rim detail, linen napkins, vintage silverware, and crystal wine glasses catch the candlelight. A perfectly roasted turkey on a ceramic platter sits at the center. The table is set on deep emerald green tablecloth. Moody, dramatic lighting with strong shadows, photographed in the style of Donna Hay food photography, rich jewel tones, professional food styling.Be sure to try these prompts out and send us your favorite generations!!
Until we Type Again…
FYI: There will be no newsletter this Thursday due to Christmas holidays, have a great one!
Thank you for reading yet another edition of the Digestibly Newsletter!
Reyhan
Early AI leader, advises Fortune 500 companies on AI Development, LLMOps, AI strategy, speaks (for fun) on practical AI. Turns cutting-edge theory into workflows teams can ship today.
Kevin
Ex-AWS SageMaker/Bedrock lead. Shipped infra powering 1M+ devices and $2.5M ARR. Obsessed with high-throughput, low-latency systems— and brings that discipline to every Digestibly release.