In mid-to-late July, several tech giants, including Figma, GitHub, and Google Labs, as well as AI start-ups such as Trickle and Mocha, simultaneously launched AI code products. These offerings emphasise an all-in-one experience, from ideation and development to deployment, lowering the technical barriers so that ordinary, non-technical users can also harness AI CODE capabilities to build real websites and applications, rather than just static pages, without writing code.

When asked about the difference between these products and platforms such as Lovable, Bolt, or Replit, Mocha’s founder, Nicholas Charriere, summed it up best: “build for the 99%, not just technical people”.

1. News Timeline

· 2025-07-25

Mocha (Axilla, Inc)

Nicholas Charriere: After 2 years of building, Mocha 1.0 is here. Not another mockup builder. Not another Supabase wrapper. This is what happens when you rethink app creation from first principles for people who don’t code.
https://x.com/nichochar/status/1948478152210350530

———————————————————————–
Website: https://getmocha.com
PH: https://www.producthunt.com/products/mocha-3
YC: https://www.ycombinator.com/companies/mocha
X: @get_mocha

· 2025-07-24

Opal(Google Labs)

Introducing Opal: describe, create, and share your AI mini-apps
It’s never been easier to harness the power of AI models, prompts, and tools into working AI applications. We’re excited to announce Opal, a new experimental tool from Google Labs that lets you build and share powerful AI mini apps that chain together prompts, models, and tools — all using simple natural language and visual editing…
https://developers.googleblog.com/en/introducing-opal/

———————————————————————–
Website: https://opal.withgoogle.com

Figma Make

Figma Make and AI features are moving out of beta
Starting today, Figma Make and all Figma AI features, including Code Layers, Rename Layers, Make/Edit Image, etc. are now in general availability and Figma Make is available for everyone to try. Access varies by plan and seat type…
https://www.figma.com/release-notes/?title=make-ai-out-of-beta

———————————————————————–
Website: https://www.figma.com/make/

· 2025-07-23

Trickle 2.0

Introducing Magic Canvas, the world’s first agentic canvas where you create with AI in a visual flow that feels natural and alive.
https://x.com/Trickle_HQ/status/1947917675893821824

———————————————————————–
Website: https://trickle.so
PH: https://www.producthunt.com/products/trickle-3
X: @Trickle_HQ

Github Spark

GitHub Spark in public preview for Copilot Pro+ subscribers
Build and ship full-stack intelligent apps using natural language with access to the full power of the GitHub platform—no setup, no configuration, and no headaches.
https://github.blog/changelog/2025-07-23-github-spark-in-public-preview-for-copilot-pro-subscribers/

———————————————————————–
Website: https://github.com/spark

Figma Make

New updates in Figma Make
We’ve introduced a few updates to Figma Make to make editing, previewing, and iterating even smoother.…
https://www.figma.com/release-notes/?title=new-updates-figma-make

———————————————————————–
Website: https://www.figma.com/make/

· 2025-07-19

Minimax Agent

Today, we’re officially launching MiniMax Agent: a general intelligent agent built to tackle long-horizon, complex tasks…
https://x.com/MiniMax__AI/status/1935496799726543233

———————————————————————–
Website: https://agent.minimax.io/

---

2. Product Features

Most products use Claude by Anthropic for AI, React for frontend, and Supabase as the main database.

			Mocha 1.0	Trickle 2.0	Figma Make	Github Spark	HeybossAI	Minimax Agent
AI Models			Claude	Claude	Claude	Claude	OpenAI, Minimax…	Minimax
Plans & Pricing	Free		Free · 120 credits · 1 apps	Free · $0/mo · 350 credits/mo	Free Limited credits	/	Free · $0/mo · 30 credits/mo	Free · 1,000 credits
	Plan I (Monthly)		Bronze · $20/mo · 1,500 credits/mo · 5 apps	Pro · $20/mo · 1,500 credits/mo · 350 free credits	Pro · $16/mo · 3,000 credits/mo	/	Basic · $24.99/mo · 200 credits/mo	Basic · $39.00/mo · 5,000 credits/mo
	Plan II (Monthly)		Silver · $50/mo · 4,500 credits/mo ·15 apps	Premium · $50/mo · 3,750 credits/mo · 350 free credits	/	Copilot Pro+ · $35 /mo · 375 Spark msgs/mo	Premium · $49.99/mo · 500 credits/mo	Pro · $119.00/mo · 20,000 credits/mo
	Plan III (Monthly)		Gold · $200/mo · 25,000 credits/mo · 25 apps	/	/	/	/	/
Input/Prompt	prompt		✅	✅	✅	✅	✅	✅
	Figma Frame		/	/	✅	/	/	/
	Assets(image,video…)		✅	✅	✅	✅	✅	✅
	Rules		/	✅	/	/	/	/
	Knowledge		✅	✅	/	/	/	/
	MCP/API		✅	/	/	/	/	✅
	AI Tools		/	/	/	/	✅	/
	Data		/	✅	/	✅	/	/
Processing	UX	THEME	/	/	/	✅	/	/
		Mobile/Desktop	✅	✅	✅	✅	✅	/
		Direct Edit	✅	✅	✅	/	✅	✅
	Code	Edit Online	✅	✅	✅	✅ (codespace)	/	/
		Download	✅	✅	✅	✅ (github)	✅ for paid user	✅
	Database		✅	✅	✅		✅	✅
	Versions		✅	✅	✅	✅	✅	✅
	Credit Usage History		✅ Credits/conversation	✅ Credits/conversation	/	/	✅ Credits/Day	✅ Credits/Chat
Publish/Deploy	One-Click Deploy		✅	✅	✅	✅	✅	✅
	Custom Domain		✅	✅	✅	/	✅	✅
	SEO		/	✅	✅	/	✅	/
	Analytics		/	✅	/	/	✅	/

2.1 Plans & Pricing

Most AI products use a subscription-to-credits model, but companies like Figma and GitHub tie these plans to their other products, which can be confusing for non-technical or non-design users:

• Monthly/Quarterly Subscriptions: They mainly focus on “SHIP IDEAS TO Websites/Apps,” but non-technical users may only use it occasionally, like creating a simple landing page or doing some creative exploration each month. Subscribing on a monthly or quarterly basis for such plans is often not cost-effective and can feel like a waste.
• Bundled Unrelated Products: AI features are often included in Figma or GitHub’s Full Seat plans, meaning users may end up paying for features they don’t need. For those who just want specific AI functionality, this “all-in-one” approach offers low value for money.

I subscribed to GitHub Spark through the $35/month Copilot Pro+ plan, but found Spark underperformed compared to other products, so I canceled immediately. If the goal is simply AI-assisted coding, there are more focused and cost-effective options on the market.

Suggestions: At the very least, provide additional payment options:

• Like Minimax Agent, sell the AI features independently without bundling with other products.

• Like ComfyUI, allow users to purchase credits without a subscription plan..

2.2 Input/Prompt

Beyond basic prompts, many products now introduce a “thinking” and reasoning stage before generation. This helps interpret user needs and structure and standardize input. For example, HeybossAI offers an AI Team Chat, where multiple AI roles discuss the user’s idea, come up with a solution, and then proceed to generation.

At the same time, these products keep adding more features and control options to enrich and guide the AI generation process. However, this also brings a new challenge: a steeper learning curve and more complex operations for users. How can they efficiently organize and leverage these tools to get results that truly match their expectations?

More input points mean greater uncertainty. This is the reality today—how to tame AI and make it easy for ordinary users to truly harness it.

2.3 Processing

Most products share similar features but differ in implementation and user interaction. In Visual Edit (Direct Edit), some use traditional element parameter settings, allowing users to precisely control each element, while others integrate elements directly into an AI conversation, allowing edits through dialogue. However, in version management, many products take a non-linear approach that can be confusing and increase complexity for users.

2.4 Publish/Deploy

In addition to One-Click Deploy and Custom Domain, some products have started addressing operational needs by expanding features in SEO and Analytics, with optimizations for social platforms and Google Analytics integration.
For users unfamiliar with SEO, HeybossAI has also introduced a paid AUTO SEO feature to provide assistance.

---

3. Product Testing

3.1 Testing Method

A simplified tool product was created for testing, without involving AI models/API calls, data storage, or user registration/login. The focus was on basic front-end interactions and the publishing workflow.

3.1.1 Prompt

The prompt included only basic user flows and automatic layout requirements:

• Missing image export function: Does the product automatically handle this?
• Missing platform-specific image data: Can it automatically retrieve and format image data according to platform requirements?
• Missing basic website/app information (e.g., header and footer): Does it automatically populate these elements?

Design an online social media collage tool that supports optimal single-image dimensions for platforms like LinkedIn, X (formerly Twitter), and others.

Main User Flow:
- Select a platform tab (e.g., LinkedIn, X, etc.);
- Display a canvas corresponding to the maximum image size for that platform. Users can drag and drop images into the canvas, and the tool will automatically fill the template based on the image dimensions;
- Click “Randomize” to automatically switch between different templates and image layouts. Users can still drag and reposition images manually.

Template Structure:
- The smallest unit is a square;
- Depending on the platform’s canvas size and the number of uploaded images, the system will automatically assemble various rectangles (in styles similar to Metro layout);
- Borders between shapes can be customized in terms of visibility and thickness.

3.1.2 Testing Subjects

In addition to the AI Code products, GPT-5 and the open-source project Dyad were included for comparison.

Only Version 1 was tested, and subsequent bug fixes or optimizations were not considered. The test aimed to evaluate the basic capabilities of the AI Code products.

3.1.3 Product Testing Scoring Guide (5-Point Scale)

3.1.3.1 Scoring Dimensions and Weights

Dimension	Weight	Description
Core Functionality	60%	Measures whether the AI completes main functions: upload, layout, export, border settings, platform selection, etc.
AI Auto-Completion	20%	Measures whether the AI can automatically fill missing data or functions.
Workflow Usability	20%	Measures whether the entire workflow runs smoothly and whether the layout is responsive.

3.1.3.2 Sub-item Scores (Unified 5-Point Scale)

• Each sub-item has a maximum of 5 points.
• Scores are assigned based on completion: 1 (poor) to 5 (excellent).

3.1.3.3 5-Point Scoring Principles

Score	Meaning
5	Fully meets requirements; complete functionality with no defects
4	Mostly meets requirements; minor issues or room for improvement
3	Partially implemented; usable but with noticeable shortcomings
2	Barely usable; incomplete functionality or major issues
1	Almost unusable or completely missing

3.1.3.4 Total Score Calculation

* Scores are subjective and do not consider credits usage, costs, or generation time.

---

3.2 Mocha 1.0

Mocha is an AI-powered application builder which allows you to build custom software applications without coding. You can describe what you want with words or pictures, and keep iterating on your app until you’re ready to publish it with one click. Our goal is to empower people with zero technical background to build working, useful software, not just mockups.

3.2.1 WORKSPACE

3.2.2 Preview

• Responsive Layout
• Switch button: Desktop/Mobile layout
• Supports selecting text elements for editing

Direct Edit， Select Text elements to Edit

⚠️ Edit text, Not all text can be edited directly. If text is dynamically generated, Direct Edit Mode won’t work and you’ll need to ask the AI to make changes instead.

3.2.3 Code

• Download code
• Edit code online

3.2.4 Database

There are different databases for the sandbox and production.

⚠️ This project does not require user registration or login, nor does it store any data, yet the platform still automatically creates three tables by default

3.2.5 Version

Every prompt generates a new version with a timestamp.

It is difficult to identify which version in the current conversation corresponds to the published content.

⚠️ Issue: The component floats to the bottom when all panels are collapsed.

3.2.6 APIs

Supports API calls, provided the configuration is set up first in Secrets.

3.2.7 Publish

Custom domain: Available to paid users

3.2.8 Output & Testing

Version 1 online：https://dkvrtkvhf74pe.mocha.app/

Dimension	Sub-item	Max Score	Score (1-5)	Notes
Core Functionality	Header/Main/Footer	5	4	No footer
	Platform Selection	5	5
	Canvas Layout	5	2	The layout is irregular and randomly arranged.
	Image Upload	5	5
	Random Layout	5	5
	Border Settings	5	2	Add border to each image
	Export Image	5	5
AI Auto-Completion	Export Function	5	5
	Platform Size Data	5	5	Linkedin Post/X Post/Instagram Post/ Instagram Story/ Facebook Post
	Header/Footer	5	3	No footer
Workflow Usability	Complete Workflow	5	2
	Responsive Layout	5	5
Total Score: 81.33333333

---

3.3 Trickle 2.0

Trickle is an all-in-one platform that transforms how you build websites, apps, and forms. At the heart of Trickle is the Magic Canvas, a fully visual canvas where humans and AI collaborate in real time. Instead of relying on linear prompts or static outputs, Trickle introduces a more intuitive and flexible way to build by making context visible, editable, and shared.

3.3.1 WORKSPACE

3.3.2 Preview

• Responsive Layout
• Switch button: Desktop/Mobile layout
• Supports selecting text elements for editing

Direct Edit， Select Text elements to Edit

• Modify element properties directly via parameter settings
• Edit elements using AI prompts

3.3.3 Code

• Download code
• Edit code online

3.3.4 Magic Canvas

⚠️ Note: Canvas zoom in/out is not supported. Drag the canvas to view all content

Version

Clearly indicate the status of the currently active version and the published version.

Assets / Knowledge / Rules / Database

Assets: Drag in your logo, images, and other brand assets. Give them clear names so the agent can recognize and reuse them correctly across your project.

Knowledge: Drop in any reference docs, FAQs, or internal context that can help the AI better understand your business and make smarter decisions.

Rules : Specify key instructions by adding rules or notes on the canvas. These help guide the AI’s behavior, or let the agent infer them for you.

Database: If your project involves dynamic content or structured data, prompt the agent to trigger the built-in database and start defining collections and entries.

3.3.5 Publish

• Custom domain: Available for paid users
• SEO: Search Result Preview / Social Media Shared Content / X (Twitter) Shared Content

3.3.6 Analytics

3.3.7 Output & Testing

Version 1 online：https://preview__akkphqui0tnf.trickle.host/

Dimension	Sub-item	Max Score	Score (1-5)	Notes
Core Functionality	Header/Main/Footer	5	4	No footer
	Platform Selection	5	5
	Canvas Layout	5	5
	Image Upload	5	3	Limit upload to 4 images
	Random Layout	5	4	The layout options are limited
	Border Settings	5	5
	Export Image	5	1	Click button does not download.
AI Auto-Completion	Export Function	5	5
	Platform Size Data	5	5	Linkedin/X/Instagram/Facebook/Pinterest
	Header/Footer	5	3	No footer
Workflow Usability	Complete Workflow	5	1
	Responsive Layout	5	5
Total Score: 75.61904762

---

3.4 Figma Make

Make your ideas real with AI. Start with a design and prompt your way to a functional prototype, fast.

3.4.1 WORKSPACE

3.4.2 Preview

• Responsive layout
• Switch button: Desktop layout / Mobile layout
• Support: Selecting elements to edit

Direct Edit, Select Elements to Edit

• Editing tools are displayed based on the type of element selected—for example, text headings (H1, H2), div tags, or button elements.
• Quickly jump to the code block corresponding to the selected element.

3.4.3 Code

• Download code
• Edit code online

3.4.4 Version

• Difficult to identify which version in the current conversation corresponds to the published
• Need to scroll through the conversation history to locate the corresponding version label

3.4.5 Database

In Make Settings / Integrations, manually link a Supabase account and create the corresponding project database.

3.4.6 Publish

• Custom domain: Available for paid users
• SEO: Social Media Shared Content
• Analytics: Use Google Analytics ID for tracking

3.4.7 Output & Testing

Version 1 online: https://visor-oak-03785567.figma.site

Dimension	Sub-item	Max Score	Score (1-5)	Notes
Core Functionality	Header/Main/Footer	5	4	No footer
	Platform Selection	5	5
	Canvas Layout	5	3	Layout overflow
	Image Upload	5	5
	Random Layout	5	1	Click does not trigger random layout
	Border Settings	5	5
	Export Image	5	1	Click button does not download.
AI Auto-Completion	Export Function	5	5
	Platform Size Data	5	5	Linkedin/X/Instagram/Facebook
	Header/Footer	5	3	No footer
Workflow Usability	Complete Workflow	5	1
	Responsive Layout	5	5
Total Score: 70.47619048

---

3.5 Github Spark

GitHub Spark is an AI-powered tool for creating and sharing micro apps (“sparks”), which can be tailored to your exact needs and preferences, and are directly usable from your desktop and mobile devices. Without needing to write or deploy any code.

3.5.1 WORKSPACE

3.5.2 Preview

⚠️ Issue: The canvas only switches to the desktop layout when its width exceeds 1264 px, causing most users to see only the mobile view.

3.5.3 App Setting

THEME

Change the styling of your app using Spark’s built-in tools

• Typography
• Color
• Border radius / Spacing

Others

• Data: Spark detects the need to store data in your app, it will automatically set up data storage
• Prompt: Spark automatically detects when AI is needed for features in your app. It will auto-generate the prompts for each AI feature, integrate with the best-fit models, and manage API integration and LLM inference on your behalf.
• Assets: Add images, logos, videos, documents or other assets to personalize your app

3.5.4 Code

• Edit code online
• create repo / open codespace

3.5.5 Version

• No version numbers
• Difficult to identify which version in the current conversation corresponds to the published content
• Click the active node to preview or publish

3.5.6 Publish

3.5.7 Output & Testing

Version 1 online: https://socialgrid-collage-m–dongou1024.github.app/

Dimension	Sub-item	Max Score	Score (1-5)	Notes
Core Functionality	Header/Main/Footer	5	4	No footer
	Platform Selection	5	5
	Canvas Layout	5	1
	Image Upload	5	1	Unable to upload images.
	Random Layout	5	1
	Border Settings	5	1
	Export Image	5	1
AI Auto-Completion	Export Function	5	5
	Platform Size Data	5	5	Linkedin/X/Instagram/Facebook
	Header/Footer	5	3	No footer
Workflow Usability	Complete Workflow	5	1
	Responsive Layout	5	5
Total Score: 53.33333333

---

3.6 HeyBossAI

HeyBoss AI Boss Mode is praised for its ease of use and efficiency in website creation, especially for those with limited technical skills. Users appreciate its ability to generate professional, SEO-optimized sites quickly by simply describing their vision. The platform’s customization options and user-friendly interface are highlighted as major strengths. However, some users note that templates can feel repetitive and advanced features may require upgrades. Overall, it’s recommended for small business owners and independent creators seeking a straightforward, effective solution for launching online projects.

3.6.1 WORKSPACE

CHAT MODE

ChatMode introduces something revolutionary to AI coding: conversation before construction.

In the AI team, various virtual members — such as the CEO, designer, and developer — participate in project discussions and ultimately deliver a conclusion.

Two Columns

3.6.2 Preview

• Responsive layout
• Switch button: Desktop layout / Mobile layout
• Support: Selecting elements to edit

Direct Edit, Select Elements to Edit

Editing tools are displayed based on the type of element selected

3.6.3 App Tools

• APP Store: No API keys. No manual setup. No integration headaches. Just describe what you need, and our AI handles everything behind the scenes.
• Upload / Library :
  • Upload images, logo, and media files.Supports JPG, JPEG, PNG, WEBP, MP4, WEBM, MP3, WAV, AAC, M4A, PDF formats
  • Library : AI Generator, Photos, Videos , Graphics, Gifs, Stickers, Audio, Icon
• Payment: No gateways to set up. No security headaches. Just connect Stripe and start accepting payments — instantly.

3.6.4 Code

• Download requires an extra purchase beyond the subscription plan
• Online editing not supported

3.6.5 Admin dashboard / Data

• Automatically detect whether an admin dashboard and database are needed
• Manually activate the database

3.6.6 Version

• Difficult to identify which version in the current conversation corresponds to the published content
• Click the active node to preview or publish

3.6.7 Analytics

3.6.8 Publish

• Custom domain: Available for paid users
• SEO
  • Manual SEO Settings: Search Result Preview / Social Media Shared Content / X (Twitter) Shared Content /Google Analytics
  • Auto SEO: Technical SEO/ Content SEO / Social SEO

3.6.9 Output & Testing

Version 1 online: https://d4ef63d1.heyboss.tech/b0bd813b/

Dimension	Sub-item	Max Score	Score (1-5)	Notes
Core Functionality	Header/Main/Footer	5	5
	Platform Selection	5	2	Overflows and cannot be switched
	Canvas Layout	5	1
	Image Upload	5	1	Unable to upload images.
	Random Layout	5	1
	Border Settings	5	1
	Export Image	5	1	Click button does not download.
AI Auto-Completion	Export Function	5	5
	Platform Size Data	5	5	Linkedin/X
	Header/Footer	5	5
Workflow Usability	Complete Workflow	5	1
	Responsive Layout	5	5
Total Score: 52.57142857

---

3.7 Minimax Agent

MiniMax Agent is an intelligent AI partner powered by cutting-edge multimodal technology. It features MCP multi-agent collaboration that brings AI teams together to solve complex problems efficiently. It offers a wide range of functionalities including instant answers, visual analysis, immersive voice conversations, and professional writing, transforming how users work and learn. With lightning-fast document summarization and an innovative floating ball feature, MiniMax Agent aims to make complex tasks effortless and boost productivity by 10x for students, professionals, creators, and anyone seeking an AI ally for writing, translation, programming, learning, or conversation.

3.7.1 WORKSPACE

3.7.2 Preview

• Responsive layout
• Support: Selecting elements to edit

Direct Edit, Select Elements to Edit

Select an element and quickly modify it via chat.

⚠️ Issue: When selecting the latest element, the selection state of previously selected elements is not cleared.

3.7.3 MCP

MCP can be configured when starting a new task, but once the conversation begins, there is no entry point on the page to access MCP settings.

3.7.4 Code

• Supports code download
• Does not support online code editing

3.7.5 Version & Publish

• Each version is automatically published with a unique URL;
• Custom domain: Available for Pro users

3.7.6 Output & Testing

Version 1 online: https://be9y10gan5gc.space.minimax.io/

Dimension	Sub-item	Max Score	Score (1-5)	Notes
Core Functionality	Header/Main/Footer	5	5
	Platform Selection	5	3	Overflows
	Canvas Layout	5	3	Uploading over 6 images won’t display.
	Image Upload	5	5
	Random Layout	5	5
	Border Settings	5	5
	Export Image	5	5
AI Auto-Completion	Export Function	5	5
	Platform Size Data	5	5	Linkedin/X/Instagram/Facebook/Pinterest
	Header/Footer	5	5
Workflow Usability	Complete Workflow	5	4
	Responsive Layout	5	5
Total Score: 91.14285714

---

3.8 Opal

Opal, a new experimental tool from Google Labs that lets you build and share powerful AI mini apps that chain together prompts, models, and tools — all using simple natural language and visual editing. Opal is a great tool to accelerate prototyping AI ideas and workflows, demonstrate a proof of concept with a functional app, build custom AI apps to boost your productivity at work, and more. 

📣 It’s hard to understand Opal’s UX design and its new innovative Node Flow model. Just like the earlier Whisk product, Google Lab always seems to create products that disrupt user habits and raise the learning curve, making people quickly lose the patience to try them.

---

3.9 GPT-5 

Output & Testing

Version 1 online: https://chatgpt.com/canvas/shared/6895560525308191bf446830c9e3338b

Dimension	Sub-item	Max Score	Score (1-5)	Notes
Core Functionality	Header/Main/Footer	5	5
	Platform Selection	5	5
	Canvas Layout	5	2	Layout overflow ,Only 4 images are displayed.
	Image Upload	5	5
	Random Layout	5	5
	Border Settings	5	5
	Export Image	5	1	Click button does not download.
AI Auto-Completion	Export Function	5	5
	Platform Size Data	5	5	Linkedin/X/Instagram/Facebook/Pinterest /custom
	Header/Footer	5	5
Workflow Usability	Complete Workflow	5	1
	Responsive Layout	5	5
Total Score: 80

---

3.10 Dyad

Dyad is a free, local open-source AI app builder. It’s an alternative to v0, Lovable or Bolt.new, but it runs on your computer, meaning there’s no lock-in.

https://www.dyad.sh

3.10.1 Download APP

3.10.2 AI Models

• Custom models: Support for OpenAI-compatible APIs
• Local models: Support for LM Studio or Ollama

3.10.3 Version

Dyad’s versioning system is always additive, meaning it’s safe to undo at any time.

3.10.4 Token Usage

Monitor token usage per Chat

3.10.5 Publish

Step-by-step guide for deploying your app with GitHub and Vercel.

Deployment has failed
Issue: ERR_PNPM_OUTDATED_LOCKFILE  Cannot install with "frozen-lockfile" because pnpm-lock.yaml is not up to date with <ROOT>/package.json

Solution :Project settings → Build & Development Settings → Install Command 
change it to: pnpm install --no-frozen-lockfile

3.10.6 Output & Testing

3.10.6.1 AI Model : qwen3-coder-plus-2025-07-22

Version 1 online: https://vibrant-hummingbird-wag.vercel.app/

Dimension	Sub-item	Max Score	Score (1-5)	Notes
Core Functionality	Header/Main/Footer	5	5
	Platform Selection	5	5
	Canvas Layout	5	2	Layout breaks when more than 4 images
	Image Upload	5	5
	Random Layout	5	5
	Border Settings	5	5
	Export Image	5	1	Click button does not download.
AI Auto-Completion	Export Function	5	5
	Platform Size Data	5	5	Linkedin/X/Instagram/Facebook
	Header/Footer	5	5
Workflow Usability	Complete Workflow	5	1
	Responsive Layout	5	5
Total Score: 80

3.10.6.2 AI Model : kimi-k2-0711-preview

Version 1 online: https://cozy-hedgehog-dive.vercel.app/

Dimension	Sub-item	Max Score	Score (1-5)	Notes
Core Functionality	Header/Main/Footer	5	4	No footer
	Platform Selection	5	3	Overflows
	Canvas Layout	5	2	Layout breaks when more than 6 images
	Image Upload	5	5
	Random Layout	5	1	Click does not trigger random layout
	Border Settings	5	5
	Export Image	5	1	Click button does not download.
AI Auto-Completion	Export Function	5	5
	Platform Size Data	5	5	Linkedin/X/Instagram/Facebook/Pinterest
	Header/Footer	5	3	No footer
Workflow Usability	Complete Workflow	5	1
	Responsive Layout	5	5
Total Score: 65.33333333

Summary

Rank	Product	Score
1	Minimax Agent	91.14285714
2	Mocha 1.0	81.33333333
3	OpenAI GPT-5	80
4	Dyad + qwen3-coder-plus-2025-07-22	80
5	Trickle 2.0	75.61904762
6	Figma Make	70.47619048
7	Dyad + kimi-k2-0711-preview	65.33333333
8	Github Spark	53.33333333
9	HeyBossAI	52.57142857

• This score is a subjective evaluation based on Version 1 outputs generated by each AI product and does not reflect their actual capabilities.
• An additional “Non-Technical User Friendliness” metric is included, rated subjectively based on learning difficulty and overall user experience.

Startup products often deeply integrate large AI models, whether by optimizing model capabilities through prompt engineering or by adding auxiliary tools to improve workflows, with the goal of enhancing product completeness and meeting user expectations. However, this approach also brings certain limitations, making products relatively passive:

• Unstable AI model core: Models are continuously evolving, and products deeply tied to a specific model tend to passively follow changes, requiring frequent adjustments.
• Closed ecosystem: Auxiliary tools are usually deeply customized around their own products, and extending their capabilities is often constrained by development resources and technical capacity.

Currently, these products are becoming increasingly similar in functionality, and vendors are constantly updating them to fill gaps. Yet, they remain vulnerable to open-source alternatives. For example, during my trial of online products, I discovered the Dyad project: it offers free access, multiple model options, and relatively complete functionality. Although using it still requires some technical knowledge (e.g., understanding model APIs, GitHub, and Vercel), its clear, step-by-step documentation significantly lowers the learning curve.

For me personally, if using an online product requires both time and money, I would rather use an open-source tool. My personal time limit is one hour: if the free quota does not yield satisfactory results and subsequent use requires a subscription to achieve uncertain outcomes, it is more efficient to give up early.

For established companies like Figma and GitHub, AI Code products are more of an exploration—providing an additional AI experience for existing subscribers rather than serving as an ideal standalone solution.

In the future, I may continue using Mocha, Trickle, and the open-source Dyad