Chat, API and browser workspace

Frontier models, one subscription

Chat with Claude and Gemini models, call them from your own code through an OpenAI- or Anthropic-compatible API, and write code in a VS Code-style workspace that runs in your browser tab.

No free tier. Plans start at $8 a month.

Claude Haiku 1x

Summarise this diff and tell me what could break.

Two things to check:

  • The retry path can run twice.
  • The limit is read before the lock is taken.
Ask anything

Models

  • Kimi
  • Claude
  • Gemini
  • GPT-OSS

Four things, one account

01

Chat across models

Switch models in the middle of a conversation. Every plan includes the full catalogue.

02

One key, two dialects

Use OpenAI-compatible or Anthropic-compatible endpoints with the same Abstract API key.

03

Evilcoder personas

Attach a coding persona to any chat. Each one sets its own instructions and tone.

04

VS Code in your browser

Dusk runs a Node-compatible runtime, a shell and a terminal inside the tab. Your files stay in your browser.

Included on every plan

Try it in the browser

Open a VS Code-style workspace right here. Edit files, run Node scripts and use the terminal without installing anything.

  • Included on every plan.
  • Files persist in this browser.
  • Loads only when you open it.
server.js

Abstract AI playground

Open the playground

Loads only when you open it. Files stay on your device.

Loading the workbench…

One endpoint, your SDK stays the same

Point the OpenAI or Anthropic SDK at Abstract and change the base URL and key. Streaming works the same way.

curl https://chat.abstracteducation.loan/v1/chat/completions \
  -H "Authorization: Bearer $ABSTRACT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-haiku-5-5", "messages": [{"role": "user", "content": "Hello"}]}'
  • OpenAI SDK
  • Anthropic SDK

Plans

Every plan includes everything. Plans differ only in how much you can use in a 5-hour window and in a week.

Billing period

Starter

To try it properly

$8$6per month

billed annually

5-hour
5,000 weighted tokens
Weekly
50,000 weighted tokens
Choose Starter

Pro

For daily work

$44$35per month

billed annually

5-hour
75,000 weighted tokens
Weekly
750,000 weighted tokens
Choose Pro

Max

For heavy use

$109$89per month

billed annually

5-hour
200,000 weighted tokens
Weekly
2,000,000 weighted tokens
Choose Max

Ultra

For the heaviest use

$219$179per month

billed annually

5-hour
500,000 weighted tokens
Weekly
5,000,000 weighted tokens
Choose Ultra

Included on every plan

Everything in the list below

  • Every model in the catalogue
  • OpenAI-compatible and Anthropic-compatible API access
  • Evilcoder personas
  • Browser workspace (Dusk playground)
  • Deep research
  • Live usage meters and reset times
  • Chat history and API keys in one dashboard

How usage is counted

Each reply is measured in weighted tokens: its token count times the multiplier of the model it ran on. A heavier model uses more weighted tokens for the same reply.

Two rolling windows: 5 hours and one week. Usage ages out as time passes, so you never wait for midnight.

Multipliers apply per model family. Versions within a family cost the same.
Model tierMultiplierA 1,000-token reply costs
Efficient models1x1,000 weighted tokens
Fast models2x2,000 weighted tokens
Balanced models3x3,000 weighted tokens
Frontier models10x10,000 weighted tokens

Common questions

Is there a free plan?

No. The lowest plan is Starter at $8 a month, and it includes every feature with a smaller usage window.

What happens when I hit a limit?

Requests pause until the window frees up. The dashboard shows how long until each window resets. You can switch to a lighter model or upgrade at any time.

What is a usage window?

There are two rolling windows, 5 hours and one week. Old usage drops out as time passes, so a limit you hit at 2pm clears by 7pm.

Can I use the API from my own code?

Yes. Create a key in the dashboard and point your SDK at the Abstract base URL. Each call counts against the same windows as chat.

What is Evilcoder?

Evilcoder is a set of coding personas. Each one has its own instructions, tone and focus, and you can attach one to any chat or playground session.

Do you offer refunds?

[PLACEHOLDER] Refund policy not yet written. The final policy replaces this answer before launch.

Start with Starter

Every plan includes every model and feature.