Deno Desktop vs Electron
Last updated 2026-09-25What's new
- Claude Opus 5.5 (a new AI model from Anthropic) is faster, cheaper, and outperforms previous models in many tasks, like coding and knowledge work (tasks like writing emails or creating PowerPoint presentations).
- It excels in coding tests, scoring 66.4% on Terminal Bench 4.0, a significant improvement over other models.
- Opus 5.5 also performs well in real-world knowledge work tasks, with a high score of 1846 on GDP val version 2.1, a benchmark for testing skills like data entry and word processing.
- The model is 30% faster and has a 20% price reduction compared to its predecessor, Opus 5, making it a more efficient and cost-effective choice.
- DeepSeek, a new AI model, is 88 times cheaper than GPT Astra (another AI model) and performs similarly to GPT 5.6 Soul (a top AI model), making it a cost-effective option for various tasks.
- DeepSeek can be integrated with GPT Astra (a powerful AI model) and used within the Codeex framework (a coding environment) to delegate work, enhancing productivity and efficiency.
- DeepSeek excels in automation, software engineering, and terminal tasks, outperforming other models like Opus 5 and GPT 5.6 in benchmarks, making it a strong choice for these specific applications.
- DeepSeek can be connected to Higfield (an image and video generation tool) to create presentations and visuals, expanding its capabilities and making it a versatile tool for design and business workflows.
- SpaceX AI is close to releasing Grock 4.7 (a powerful AI model), which could be a strong competitor to existing models like Fable 5.1, with rumors suggesting it has 2.1 trillion parameters (a measure of its complexity).
- OpenAI (a leading AI company) has increased access to their new model, GBT6 Astra (a type of AI software), by resetting user limits, allowing more people to try it out.
- Anthropic (another AI company) may soon release a new model, possibly called Fable 5.2, in response to OpenAI's GBT6 Astra, to regain market share.
- Google's Gemini 4 Pro (an AI model) might be delayed again as it aims to compete with advanced models like Fable 5.1 and GBT6 Astra.
- Omachi is a new, free, and open-source (software anyone can use, modify, and share) AI-powered operating system (the main software that runs your computer) that works on most computers and laptops.
- It's designed to be super fast, lightweight, and productive, with built-in AI tools (like ChatGPT and Claude) that work together seamlessly.
- Since it's open-source, users can edit and customize any part of the operating system using AI, making it highly personalized and adaptable.
- The creator believes open-source operating systems are the future, as they can be continuously improved and tailored to individual needs.
- Grok, a new AI player, released Grok 4.6 (an AI model for coding and knowledge work), which is now competitive with top models from OpenAI and Anthropic (two leading AI companies).
- Grok 4.6 improved significantly in speed and performance, especially in coding tasks, thanks to Grok's acquisition of Cursor (a company that helped Grok enhance its AI model).
- Grok 4.6 became more expensive but also more intelligent, making it a strong competitor in the AI market, though it's still pricier than some other models.
- Grok 4.6 performed exceptionally well in legal use cases, showing a significant improvement over its previous version and other models.
- Prime Intellect (a company) is working to make advanced AI models (the brains behind AI tools) and the tools to train them open and accessible to everyone, not just big companies.
- RCAI (a company) is focusing on creating custom AI models (AI tools tailored to specific tasks) that are open, permissive (can be used freely), and can be owned and run by anyone, anywhere.
- NVIDIA's Neotron (a family of AI models) aims to be as open as possible, believing that open AI development helps everyone build and improve AI tools together.
- Neotron also focuses on making AI models faster, which is crucial for local AI (AI tools that run on your own devices, not just in distant data centers).
- Companies often overspend on "inference" (using AI to process and understand data) from services like Anthropic, as costs can quickly spiral out of control, similar to overspending in a casino.
- Some users are turning to "token factories" (using open-source AI models on your own hardware) to avoid high costs from services like Anthropic and OpenAI, but this can have reliability and control issues for businesses.
- Enterprises often face additional challenges with rented AI services, such as lack of control over usage limits, audit issues with third-party vendors, and inability to recreate AI-generated recommendations.
- The speaker recommends that startups should rent AI services until they find product-market fit, but eventually, they should build their own "inference infrastructure" (AI processing setup) to have more control and avoid ongoing costs.
- Primed and Loaded has introduced new open-source tools, Verifiers and Primed RL libraries, to help improve AI models after they're initially trained (post-training).
- They've created a global marketplace of data centers with over 10,000 GPUs (powerful computer chips) to support large-scale AI model training.
- The company is working on making it easier for anyone to train and customize AI models for their specific needs, not just using existing models.
- They've also introduced a new platform called Lab, which combines various tools to make AI research and model training more accessible and efficient.
- ChatGPT has a new feature called ChatGPT Work (previously Codex), which lets you assign tasks and get finished work like spreadsheets or reports without constant supervision.
- ChatGPT Work is now available as a desktop app (Mac and Windows) and on the web, allowing AI to access and use files and folders on your computer for better task completion.
- To use ChatGPT Work effectively, create dedicated folders for specific tasks to avoid overwhelming the AI with unrelated files, and choose the right AI model (Sol for complex tasks, Luna for simple searches).
- ChatGPT Work offers different effort levels (light to ultra) and speed options (standard or fast) to tailor the AI's performance to your task's complexity and urgency.
Key points
What it is
- **Deno Desktop** is a tool for building desktop apps using web technologies, but it uses the system's built-in browser engine (like WebView2 on Windows or WebKit on Mac/Linux) to keep the app size small (around 77 MB).
- **Electron** is another tool for building desktop apps using web technologies, but it bundles a full copy of the Chromium browser (the engine that powers Chrome) into every app, making it much larger (around 248 MB).
- Deno Desktop is experimental and less polished, while Electron is more mature and widely used.
How to use it
- To use Deno Desktop, you point it at a TypeScript file, and it serves your front-end interface into the built-in webview, with the back-end logic running in Deno.
- Deno Desktop communicates between the front end and back end through a binding, which is a direct communication channel, while Electron uses a separate channel called IPC (inter-process communication).
- Deno Desktop also includes production features like system tray API, native notifications, auto-update, and cross-compilation out of the box.
Watch out for
- Deno Desktop is experimental and may have quirks, especially on Windows, and it doesn't yet match Electron's mature ecosystem.
- If you need consistent rendering across all operating systems with Deno Desktop, you can bundle a real Chromium, but that increases the file size.
- A best practice is to abstract your IPC APIs to fall back to web standards, making your app render-agnostic and easier to test.
Tools named
- Deno Desktop (a lightweight tool for building desktop apps using web technologies), Electron (a framework for building desktop apps that bundles Chromium and Node.js), Electron (built on Bun instead of Chromium, produces smaller starter apps but is currently Mac and Linux only).
Lesson 1: What is Deno Desktop vs Electron and why it matters
Deno Desktop and Electron are both tools for building desktop apps using web technologies, but they differ in how they render the interface. Electron (a framework that bundles a full Chromium browser into every app) ships an entire copy of Chrome with your application, resulting in large file sizes—248 MB for a starter app. Deno Desktop takes a lighter approach by using the system web view (the browser engine already installed on your operating system—WebView2 on Windows, WebKit on Mac/Linux), cutting your app size to around 77 MB.
For AI development, this difference matters because AI tools often do heavy lifting on the user’s machine—reading files, launching other processes, or running local models. A smaller footprint means faster downloads and less memory overhead. Deno Desktop also gives your front-end code direct access to system APIs (functions that interact with the computer) through a binding (a direct communication channel between your web page and the back-end), skipping Electron’s more complex IPC (inter-process communication) channel. This simplifies code and speeds up interactions.
However, Deno Desktop is experimental and lacks some polish. On Windows, you may encounter web view tool quirks, and it doesn’t yet match Electron’s mature ecosystem. For AI apps that need consistent rendering across all operating systems, you can still bundle a real Chromium with Deno Desktop using one flag—but that increases file size. Electron remains more battle-tested, while Deno Desktop offers a leaner, faster path for AI developers who want immediate system access without the weight.
Sources
- 2026-07-05 — 77 MB vs 248 MB Why Deno Desktop Is Crushing Electron
- 2026-05-04 — The Exact Moment Claude Cowork Makes More Sense Than ChatGPT
- 2026-05-10 — Feedback Loops are All You Need Mehedi Hassan, Granola
- 2026-05-14 — Build an Execution Layer for Your Company Brain (Step by Step)
- 2026-02-25 — Why GPUs Are About to Lose AI Inference!
- 2026-05-06 — Master 97% of Codex in 1 Hour (full course)
Lesson 2: How to use Deno Desktop vs Electron: step-by-step
Deno Desktop and Electron both let you build desktop apps by displaying a web page alongside code that can interact with the user’s machine. The key difference is that Electron ships a full copy of the Chromium browser inside every app, which makes a starter app around 248 megabytes. Deno Desktop instead borrows the system webview (the browser engine already in your operating system: WebView2 on Windows, WebKit on Mac and Linux), so the same app compresses to roughly 77 megabytes.
To use Deno Desktop, you point it at a TypeScript file. It serves your front-end interface into that built-in webview, and your back-end logic runs in Deno. The front end and back end communicate through a binding: you call a function on the front end, and it runs on the back end in the same process. Electron requires a separate channel called IPC for this. Deno Desktop also ships production features out of the box: a system tray API for icons and menus, native notifications and dialogs, auto-update with tiny binary diff patches, and cross-compilation from one machine to target Mac, Windows, and Linux (producing app images, DMGs, EXE or MSI installers, deb, and RPM).
However, Deno Desktop is experimental and version one. On Windows, you may inherit webview tool quirks, and while the auto-detect finds nine web frameworks, you must name the entry point yourself for a raw single-file app. Electron is a more mature platform, and an alternative called Electron (built on Bun instead of Chromium) produces starter apps around 14 megabytes, but it is currently Mac and Linux only with no Windows support. Startup time is roughly equal between Deno Desktop and Electron—around 370 milliseconds—so the main win for Deno Desktop is disk size and memory usage, not launch speed. You can also pass one flag to bundle a real Chromium with Deno Desktop if you need consistent rendering across operating systems, at the cost of a larger file.
Sources
- 2026-07-05 — 77 MB vs 248 MB Why Deno Desktop Is Crushing Electron
- 2026-05-10 — Feedback Loops are All You Need Mehedi Hassan, Granola
- 2026-06-01 — 20 days of compute vs 7 hours rethinking what state-of-the-art means Bertrand Charpentier, Pruna
- 2026-07-01 — Building a Real AI Jarvis (GPT Realtime 2 + Cursor)
Lesson 3: Best practices and pitfalls
Deno Desktop and Electron both let you build desktop apps with web technology, but they differ in size, complexity, and maturity. Electron (a framework that bundles Chromium and Node.js) ships an entire browser inside every app, which makes a starter app around 248 megabytes. Deno Desktop uses your operating system's built-in webview (the browser engine already on your machine)—WebView2 on Windows, WebKit on Mac and Linux—so a similar app compresses to just 20 megabytes. That size gap is almost entirely the bundled Chromium that Electron requires.
A common pitfall with Deno Desktop is that it's experimental. It auto-detects nine web frameworks, but if you point it at a raw single-file app, you must name the entry point yourself. On Windows, you inherit webview tool quirks. Also, while Deno Desktop gives you built-in system tray APIs, native notifications, auto-update with tiny binary diff patches (kilobyte-sized updates), and cross-compilation from one machine to Mac, Windows, and Linux, Electron currently has broader platform support—Deno Desktop is still version one, not a finished platform. Startup speed is roughly the same for both (about 370 milliseconds), so the win is in disk size and memory, not launch speed.
A best practice is to abstract your IPC APIs (inter-process communication channels that connect the front-end to the system) to fall back to web standards. This makes your app render-agnostic, meaning you can run it as a web app without needing Electron or Deno. That lets you test features in parallel without installing dependencies or asking coworkers to run a local copy. If you need consistent rendering across operating systems, Deno Desktop offers one flag to bundle a real Chromium anyway—bigger file, but consistent everywhere.
Sources
- 2026-07-05 — 77 MB vs 248 MB Why Deno Desktop Is Crushing Electron
- 2026-05-10 — Feedback Loops are All You Need Mehedi Hassan, Granola
- 2026-07-01 — Building a Real AI Jarvis (GPT Realtime 2 + Cursor)
- 2026-06-01 — 20 days of compute vs 7 hours rethinking what state-of-the-art means Bertrand Charpentier, Pruna