Blog
News, research & updates
Product releases, technical deep-dives, and architectural research from the Dropstone team.
How Dropstone Became Real-World AI
Dropstone started as a coding agent. Over the past several months it learned to draft and send your email, place real phone calls on your behalf, take calls back, watch your inbox while you're away, and control the devices in your home. Here's the shift, what we call it, and why.
Dropstone 1.6 Models
Dropstone 1.6 moves to GLM-5.2, Kimi K2.7 Code, and DeepSeek V4 Flash across its three tiers. GLM-5.2 becomes the first open-weight model we've measured above 80% on Terminal-Bench 2.1, at roughly one sixth of the output cost of comparable closed models. Claude's frontier models still lead, and we say so.
Introducing Dropstone 1.5: Frontier Coding Intelligence, For Everyone
Dropstone 1.5 outperforms Claude Opus 4.7, GPT-5.5 Pro, and Gemini 3.1 Pro on real-world coding benchmarks, runs on US-hosted infrastructure, and supports a full-time coding workload at every tier. 13.9 million tokens processed in the first 12 hours of soft launch, with zero marketing.
The Desktop Becomes the Platform
v3.0.7 extends the D3 Engine beyond the editor. Windows MCP V2 provides native OS-level automation. Figma MCP V2 unifies design and engineering workflows. Over 1,000 MCP servers deliver broad ecosystem connectivity. PIP Mode introduces persistent desktop orchestration.
Share Chat: The End of Single-Player Coding
Introducing Share Chat, Team Memory, and collaborative workspaces — the features that make Dropstone the first truly multiplayer AI coding tool.
Horizon Mode: 24-Hour Reasoning Horizons
Breaking the Linearity Barrier with Recursive Swarm Architecture. How we enable coherent AI reasoning over extended time horizons.
The D3 Engine: Neuro-Symbolic Runtime Architecture
Introducing Dynamic Distillation & Deployment - how we reduce compute costs by 99% while maintaining engineering-grade accuracy.
Infinite Context Through State Virtualization
How we achieved a 50:1 compression ratio while maintaining 99.9% recall accuracy. A deep dive into our memory architecture.
AST Parsing at Scale: Tree-sitter Across 40 Languages
How Dropstone achieves deep code understanding through intelligent AST parsing. A technical deep-dive into our polyglot parsing infrastructure.
Fighting the Lag: Designing a Zero-Overhead Runtime
LLMs are inherently slow. The runtime wrapping them shouldn't be. How we are migrating the D3 Engine to Rust to strip away every microsecond of overhead from the reasoning loop.
Introducing Collaborative Sessions
We are introducing Session Forking. You can now generate immutable, shareable links to your Dropstone environments, allowing teammates to fork your exact reasoning context and continue the architectural thread.