AI Specification Generation — Model Comparison
We pointed SPECLAN's “Infer Specs from Code” agentic pipeline at the famous excalidraw codebase and asked a range of frontier LLMs — plus a few open-weights models — to reverse-engineer the product's specification from source. The same pipeline is what turns a legacy codebase into a living, AI-augmented spec tree you can actually continue to build on — spec-driven development, retrofitted.
A note on fairness: the comparison isn't apples-to-apples. Under the hood the OpenAI SDK (used for OpenAI and Local LLM runs) calls only the MCP tools we expose, while the Anthropic SDK layers in its own built-ins — Todo-Lists and friends — that measurably amplify the model's agentic strength. Read the trees with that in mind.
Claude Opus 4.7 (1M context)
Methodology notes
Generated against the excalidraw repository via the Claude Agent SDK with MCP explore tools + speclan tools. Used the extended 1M context window so the full codebase could be reasoned about in one pass. No curator touch-ups applied.
Reference baseline for this comparison. 16 top-level features map 1:1 to excalidraw's real surface — shape drawing, canvas navigation, real-time collab, text-to-diagram, command palette, undo, view modes, and more. 197 requirements with consistent acceptance-criteria rigor, and the sub-feature nesting reads like a staff engineer's carve-up. Every other tree on this page is measured against this one.
Vision: A World Where Ideas Flow Freely Through Visual Thinking
The Future We're Building
We envision a world where anyone — regardless of artistic skill, language, or technical background — can instantly translate their thoughts into clear, beautiful visual diagrams. Excalidraw exists to make visual communication as natural and effortless as speaking, removing every barrier between an idea and its expression on a shared canvas.
Key Outcomes
- Thinking made visible: Every person can sketch ideas, explain concepts, and tell stories visually — with a hand-drawn style that feels approachable rather than intimidating
- Seamless collaboration: Teams ideate together in real time, no matter where they are, with the confidence that their shared work remains private and secure through end-to-end encryption
- An open ecosystem for visual tools: Developers everywhere can embed and extend a world-class drawing experience inside their own products, creating a universal standard for visual expression on the web
- Ideas preserved and portable: Drawings live in an open format that users truly own — exportable, shareable, and never locked into a single platform
Who Benefits
Individual thinkers gain a frictionless space to brainstorm, wireframe, and diagram — one that works offline, saves locally, and asks for no sign-up. Teams and organizations unlock real-time visual collaboration that keeps sensitive information encrypted end-to-end. Developers and product builders integrate a proven, customizable drawing canvas into their own applications, from knowledge management tools and code editors to cloud platforms and educational products. The global community benefits from an open-source project available in dozens of languages, welcoming contributors and users from every corner of the world.
The Impact We Seek
When visual communication is universally accessible, understanding deepens, alignment happens faster, and creativity thrives. Excalidraw aims to be the foundational layer that makes this possible — everywhere ideas need to be seen.
GPT 5.4
Methodology notes
Generated against the excalidraw repository via the OpenAI Agents SDK with MCP explore tools + speclan tools. No curator touch-ups applied.
Compact by a factor of 4.5× in requirements (43 vs Opus's 197) and 2.7× in features (6 top-level vs 16). Each top-level bucket is a broad umbrella — 'Diagram Creation and Editing', 'Real-Time Collaboration and Sharing' — that stays at the elevator-pitch level. Good as a navigational overview, weaker as an implementation blueprint next to Opus.
Vision: Sketching Shared Understanding for Everyone
The Future We're Building
We envision a future where ideas can be explored as quickly as they are imagined. This project enables people to turn rough thoughts into diagrams, wireframes, explanations, and collaborative working sessions without friction, so visual communication feels natural instead of formal or slow.
Key Outcomes
- People can capture and shape ideas instantly on an infinite canvas.
- Teams can work together in real time without losing trust, privacy, or creative momentum.
- Visual work can be saved, reopened, shared, and reused across projects and audiences.
- Organizations can embed visual thinking into their own products and workflows.
Who Benefits
Individuals benefit from a fast, forgiving space for sketching concepts, planning flows, and explaining complex ideas. Teams benefit from a shared visual workspace that supports live collaboration, offline continuity, and simple sharing when feedback or alignment is needed. Product, design, education, and technical communities benefit from a tool that makes communication clearer, more inclusive, and more approachable through a hand-drawn style that lowers the barrier to participation.
Why It Matters
When visual thinking is easy, more people contribute, decisions form faster, and ideas become easier to understand. This project helps make collaborative problem-solving more accessible, expressive, and dependable for anyone who needs to think together visually.