Latest Updates
Hey everyone!
If you've used Crafts, you've probably hit this. You ask Merlin to build a React dashboard, a styled HTML page, or a diagram with more than a few nodes, and somewhere around the 150-line mark, the output just stops. Mid-function. Sometimes mid-tag. The AI cheerfully continues the conversation like the code is complete. It isn't.
That single bug report collected ~90 votes and had ~39 separate complaints merged into it. It was the loudest thing in our bug tracker for months, and we owe you an honest answer: the original Crafts was fundamentally broken for anything complex. We built it as a single generation call, the AI tried to output your entire file in one shot, hit its limit, and silently stopped. No retry, no warning, no signal that your code was incomplete. Just a broken component and a confused user.
So we rebuilt it from scratch.
What's new in Crafts v2
The core change: instead of one all-or-nothing generation step, the AI now builds files the way a developer would, in stages, with the ability to read what it wrote, fill in sections iteratively, and only show you the result when it's actually ready. No more silent truncation. No more half-finished code.
The result: files over 500 lines work properly. Files over 1,000 lines work. We've tested up to 7,000. The code that used to get cut off mid-function doesn't anymore.
A few other things that came with the rebuild:
- Iteration actually works. If a generated file has a bug, the AI can now read the specific section, replace just that part, and verify the fix, without regenerating the entire file from scratch.
- Model-aware generation. Claude, GPT, Gemini, DeepSeek, Kimi, and others each get instructions tuned to how they actually behave. One-size-fits-all prompting was silently failing on a lot of models. That's fixed.
What people are building with Merlin crafts:
- https://crafts.getmerlin.in/shared/2fafd160-e2e3-4eec-8478-4d13c6156cc5 - multi-section portfolio with experience timeline, project stats, and smooth animations
- https://crafts.getmerlin.in/shared/267cff8f-d4c3-4368-a4ff-d9d224a09316 - full lore breakdown across 7 nations, elements, and characters
- https://crafts.getmerlin.in/shared/20ee8c4e-4425-44c6-a387-fbb34056e7b5 - drag, compress, export as ZIP, fully functional in the browser
- https://crafts.getmerlin.in/shared/3885f359-df3c-444e-a16d-a070a73ed60b - darkroom-aesthetic photography portfolio with contact sheets and film specs
- https://crafts.getmerlin.in/shared/1bc52b2c-244f-4877-8e28-e251cad86774 - a fully playable Space Invaders-style game with upgrades, levels, and score tracking
- https://crafts.getmerlin.in/shared/9bda970c-11cd-4de6-abb0-ecb6f8c40971 - drag-to-grate mechanic with a French culinary recipe, live shred counter, and artisan typography.
Try it
The easiest test: ask Merlin to build something that would have broken before. A React component with charts and a sidebar. A styled landing page with multiple sections. A Mermaid sequence diagram with 20+ steps. That's exactly the kind of output v1 couldn't finish. It should work now.
Let us know how it goes - Shubhankit @ Merlin
We've launched a host of new models from OpenAI and xAI on Merlin.
GPT-4.1 Series
We've added three brand-new OpenAI models:
- GPT-4.1: Upgraded version of GPT-4o with better coding skills and understanding of complex instructions. Costs 15 ⚡ per non-Live response.
- GPT-4.1-mini: Similar quality to GPT-4o but faster and cheaper. Costs 3 ⚡ per non-Live response.
- GPT-4.1-nano: Replacement for GPT-4o-mini with a 100k token context window in Large Context mode. Costs 1 ⚡ per non-Live response.
Grok Models
Two new models from xAI are now available (finally):
- Grok 3: Excellent at math, reasoning, and coding problems. Costs 25 ⚡ per response.
- Grok 3 mini: Faster, cheaper version that's still great at math and coding. Costs 2 ⚡ per response.
We have plans to increase the context window on GPT-4.1 series models soon. Let's see how that goes!
Hey folks!
Watch this video to get an awesome sneak peek at some of the new stuff we're gonna release soon!
Hints: New Merlin UI, Deep Research, AI Tasks!
We've released some great stuff over this month. This video runs you through some of them:
- New UI on Merlin Extension: Same Chat experience on Merlin Extension [finally!]
- Merlin Vault and Save to Merlin: Lets you bookmark stuff when you're browsing, straight into Merlin
- Projects and Search Focus Modes on Merlin Mobile: Now you can create and access Projects and Search Focus on Merlin's mobile apps.
- Automatic Creation of Projects and Magic Import for Projects: Using AI, create well-crafted Projects and scrape relevant knowledge from the Web for your Project
- Magic Extend on Bonkers: Extend images or Bonkers generations
- Personalisation and Memory on Merlin Chat: Supply system prompt to Merlin and allow it to remember stuff about you
Share feedback if something's buggy!
Introducing three new features on getmerlin.in/chat!
Chatbots on Merlin
Create custom bots in 3 simple steps:
- Name your bot
- Define its purpose
- Set welcome message
Features:
- Public/private visibility options
- Trigger with @bot-name
- Works in extension
DeepSeek V3 Model
New open-source AI model now available on Merlin!
- GPT-4o level performance at lower cost
- 37B active parameters from 671B total
- Superior coding and math capabilities
- Multilingual support
- Cost: 2 queries per prompt
Pro Finder
Enhanced project search feature:
- Improved file searching for large projects
- Uses AI-driven validation and agentic framework
- Costs 250 queries per use for non-Pro users
- Experimental feature with optimization ongoing
Hey Merlin family! We're excited to launch Focus Mode - a smarter way to search specific types of content right from your chat!
What's New?
Three specialized search modes to help you find exactly what you need:
YouTube Mode
- Search across YouTube videos and their content
- Perfect for finding reviews, podcasts, or specific creator content
- Example: "What were MKBHD's top phone picks this year?"
Academic Mode
- Powered by Semantic Scholar for research discovery
- Great for exploring latest research with inline citations
- Example: "Show me recent research on LLM tokenization"
Social Mode
- Currently Reddit-focused for community insights
- Ideal for real-world opinions and comparisons
- Example: "Compare user experiences with Airpods Max vs XM5"
How to Use
- Find the Live Search toggle below your chat bar
- Select your desired Focus Mode
- Start searching with domain-specific context!
