Long-horizon agentic coding with 300-sub-agent swarms and 4,000 coordinated steps. Built on Kimi K2.6, hosted in the US.
The Astrophysical Journal accepted our AI's FRB discovery after three rounds of peer review, but the AAS editorial office then halted the paper over an AI-disclosure review, not the science.
Read moreWhy Blankline is working towards a 10-million agent architecture grounded in directed learning and safety.
Read moreDecoupling reasoning depth from context length via Recursive Swarm Architecture and heterogeneous inference routing.
Read moreBy decoupling probabilistic reasoning from deterministic state, D3 eliminates context saturation enabling reliable, long-horizon autonomous engineering agents.
Read moreStandard public coding benchmarks, latest published runs.
Dropstone Heavy is the tier for the work that takes a whole afternoon and a whole repository. Built on Moonshot Kimi K2.6, scoring 58.6 on SWE-Bench Pro (vs 53.4 for Claude Opus 4.6) and 54.0 on Humanity's Last Exam with tools, leading every published model. Scales horizontally to 300 sub-agents executing 4,000 coordinated steps inside the same approval-gated CLI.
Migrate a monorepo from one framework to another. Heavy plans the work, decomposes it across hundreds of sub-agents, and reports back when the build is green.
Up to 300 parallel sub-agents working on different parts of a codebase at once, coordinated through the same approval gate that protects every other Dropstone session.
Tool-augmented reasoning over scientific datasets. The same workflow that drives Blankline's FRB and wormhole analysis pipelines.
Every Dropstone request is treated as if the model could be adversarial. The CLI requires explicit user approval before any action that writes to disk, runs a shell command, or fetches a URL. No model output is ever auto-executed.
Heavy runs on SOC 2-certified, US-based inference providers. Prompts and completions are not retained, not used for training, and not shared with the model's original operator.
Heavy is built on Moonshot Kimi K2.6, an open-weight foundation model. Goldwasser et al. (2022) proved no party can prove a closed foundation model is free of embedded behaviors, including Anthropic for Claude and OpenAI for GPT. We say this out loud. The runtime is why model origin does not matter for your code.
Same approval gate, same US-hosted inference, same zero-retention guarantee across all three. Pick the smallest model that meets the task.
Sub-second agentic coding for edits, refactors, and high-throughput inline completion. Built on DeepSeek V4 Flash, hosted in the US.
Frontier agentic coding within 0.2 points of Claude Opus 4.7 on SWE-bench Verified. Built on DeepSeek V4 Pro, hosted in the US.
Long-horizon agentic coding with 300-sub-agent swarms and 4,000 coordinated steps. Built on Kimi K2.6, hosted in the US.
Fast is included on every plan. Pro and Heavy are unlocked on the Pro plan ($15/mo) and Max plan ($75/mo).
Install the CLI, authenticate, and start running approval-gated agentic workflows. No credit card to start.
