Dropstone

Pricing / Pro

More of what makes
Dropstone a useful partner.

Create, iterate, and ship faster without hitting limits on the Pro plan. Heavy-tier reasoning, the research roadmap, and the same privacy guarantees as Free.

Get startedDownload Dropstone

$15/mo · cancel anytime · no setup

Why Pro

Dropstone thinks with you.

FIG 0.1

Stay in flow

4× more usage than Free, so the loop between question and answer never breaks mid-task.

FIG 0.2

Reach the Heavy tier

The audited Heavy model unlocks at Pro. The tier we reach for on long-context and ambiguous specs.

FIG 0.3

Delegate the tedious

Context summarization with precision engineering. Saved presets and larger working context per session.

FIG 0.4

Plug into your stack

Remote MCP brings any tool into the agent. Test runners, build systems, internal APIs, all in one loop.

Built with Dropstone Pro

Wait, you can do that?

Real artifacts shipped by Heavy in one prompt. Not just code.

Prompt

Turn `research-notes.md` into an arXiv-ready LaTeX paper. Cite the routing literature, write the abstract, and compile the PDF.

Attachments

research-notes.md

4,820 lines

md

refs.bib

62 entries

bib

arXiv:2606.14728v1 [cs.LG] 11 Jun 2026

Tier-Aware Inference Routing for
Cost-Bounded Coding Agents

M. Reyes-Okafor1 · A. Lindqvist1 · K. Iwasaki1,2

1Blankline Research    2MIT CSAIL

{m.reyes, a.lindqvist, k.iwasaki}@blankline.org

Abstract

We present a tier-aware routing strategy that selects among three serving tiers (Fast, Pro, Heavy) using a learned cost–quality estimator. On 24,180 production coding requests, the policy achieves 94.2% of frontier-tier quality at 38% of the cost.

1.  Introduction

Coding agents face a routing problem at every turn: each request can be served by a small fast model or escalated to a larger one with higher cost. Static policies under-deliver on hard requests and overspend on trivial ones. We formalize tier selection as

maxπ 𝔼x∼𝒟[Q(π(x), x)] s.t. 𝔼x∼𝒟[C(π(x))] ≤ B(1)

where Q is per-request quality, C is serving cost, and B is the per-user budget. Section 2 derives the dual-form solution using a learned Lagrange multiplier [12, 18].

Research paper

Draft, format, and ship a LaTeX paper from research notes. Dropstone writes the abstract, sections, citations, and equations, then runs the build.

Why people choose Pro

06 reasons

  • 014× more usage than Free, with no surprise rate limits mid-task
  • 02Dropstone Heavy unlocked, the tier engineered for long-context and ambiguous specs
  • 03Saved agent presets and system prompts that persist across sessions
  • 04Access to Blankline Research and the latest engineering upgrades
  • 05Beta Research Program and first look at monthly audit results
  • 06Same zero-retention privacy posture as every other Dropstone tier

One plan for getting started.

Need more? Max gives you 6× usage and priority routing.

Pro

Recommended

More usage, the Heavy tier, and the research roadmap

$15 /mo
  • 4× more usage than Free
  • Dropstone Heavy with XHigh Thinking
  • Saved agent presets
  • Access to Beta Research Program
Start Pro

Max

The most usage, priority queue, and the earliest features

$75 /mo
  • 6× more usage than Pro
  • Priority routing at peak hours
  • Dedicated capacity reserved
  • Early access to new features and models
See Max

Ready when you are.

Start Pro