Try Raven

Dipole ML Inc.

Canadian frontier AI, available globally

Dipole ML Inc. is a Toronto-based Canadian frontier AI lab that builds the Raven model family, DCode, and the DipoleML API. Raven Max and Raven Flash lead public tests of terminal and agent work, with your data encrypted at rest and never sold.

Try RavenExplore DCode

Two apps, built on our own Raven models

Raven on a Mac: a sidebar with chats and projects beside a conversation planning an offsite.
DCodeOne prompt, a crew of coding agents, one verified answer.See DCode at work
RavenThe AI app that takes the whole job, on the web, desktop and phone.See everything Raven does

Frontier intelligence that leads the world in terminal execution and complex reasoning, delivered with your data encrypted at rest and never for sale.

The models

One foundation, two Raven models

Raven Max and Raven Flash share one foundation: text and image understanding, native tool calling, and reasoning you can inspect step by step.

Raven Max · Frontier reasoning

Deep intelligence for high-stakes work

Our flagship model for complex engineering, long-horizon planning and rigorous analysis. It plans across multi-step dependencies, calls tools with precision and streams its reasoning, so every decision can be audited.

Terminal Bench 4.0
42.8
Agent's Last Exam
39.5
GDPVal-AA
1771
See the benchmarks

Raven Flash · High-velocity frontier

Frontier capability at production speed

Built for low latency and high throughput without giving up accuracy. Raven Flash is the default engine in DCode and the Raven app, for real-time answers, tool use and code.

AutomationBench
49.2
Input, per 1M tokens
$0.50
Output, per 1M tokens
$2.50
See the API

Benchmarks

Measured against the frontier

Raven models are built for terminal work, autonomous tool use and long multi-step reasoning. These are the public tests we report.

42.8Raven Max on Terminal Bench 4.0, the top score among the models compared.
Raven MaxBenchmark leader
42.8
Fable 5
42.4
GPT-5.6 Sol
37.9
Muse Spark 1.3
33.3
Gemini 3.8 Flash
19.7
Terminal Bench 4.0
ModelScore
Raven Max42.8
Fable 542.4
GPT-5.6 Sol37.9
Muse Spark 1.333.3
Gemini 3.8 Flash19.7
Agent's Last Exam
ModelScore
Raven Max39.5
Fable 531.6
GPT-5.6 Sol29.8
Kimi K327.6
Gemini 3.8 Flash24.7
AutomationBench
ModelScore
Raven Flash49.2
Fable 548.6
Kimi K346.7
GPT-5.6 Sol44.0
Gemini 3.8 Flash43.5
GDPVal-AA
ModelScore
Raven Max1771
Fable 51760
GPT-5.6 Sol1730
Muse Spark 1.31710
Gemini 3.8 Flash1647

Evaluated with maximum reasoning effort enabled. Verified against official frontier model evaluations.

Raven app

Hand Raven the whole job

Ask anything and it answers. Give it a job and it opens a browser, reads the web, writes the report and makes the image. One account on the web, desktop and phone.

  • Browses for youFinds the cheapest direct flight in a real browser, one step at a time.
  • Researches with sourcesPlans the work, reads the web and writes a report you can check.
  • Makes the pictureTurns one sentence into an image, right inside the chat.
See everything Raven doesOpen Raven on the web
Raven on a Mac: a sidebar with chats and projects beside a conversation planning an offsite.

DCode

One prompt, a crew of agents, one verified answer

  1. 1

    Start with one sentence

    Type the task the way you would say it to a teammate. There are no tickets to split and no agents to configure.

  2. 2

    DCode splits the work

    A crew lead reads your repository, breaks the task into independent parts and writes a shared brief, so names and interfaces stay consistent.

  3. 3

    The crew works in parallel

    Each agent takes a part. The Crew dock shows who is doing what and who needs you. Answer a question once and every agent asking it gets the answer.

  4. 4

    Checked before it is reported

    A reviewer runs your real build and tests, then merges the parts. You get one answer that says exactly what was verified.

DCode 2.5.0 is out for Windows and macOS, with a CLI for your terminal.

See DCode at workDownload DCode

DipoleML API

Change one line and your tools run on Raven

The API speaks both the OpenAI and Anthropic formats. Point the SDK you already use at api.dipoleml.com and keep the code you have.

  • Streaming replies, tool calling and image input
  • Prompt caching discounts of up to 90%
  • Pay as you go, with prepaid credits that never expire
Read the API docsOpen the playground

Python

from openai import OpenAI# Drop-in replacement: only the base_url changesclient = OpenAI(base_url="https://api.dipoleml.com/v1",api_key="your-dipole-api-key")response = client.chat.completions.create(model="Raven Max",messages=[{"role": "user", "content": "Refactor the auth microservice."}])print(response.choices[0].message.content)

cURL

curl https://api.dipoleml.com/v1/chat/completions \-H "Authorization: Bearer your-dipole-api-key" \-H "Content-Type: application/json" \-d '{"model": "Raven Max","messages": [{"role": "user", "content": "Refactor the auth microservice."}]}'
Raven Max

I would split token checks out of the request handler first, then add a refresh path with tests. I can open a pull request with the three changes.

Works with the coding apps, harnesses and ADEs engineers already use

  • Claude Code
  • Cursor
  • Codex
  • Windsurf
  • Cline
  • Kilo Code
  • pi
  • OpenCode
  • JetBrains
  • Trae
  • OpenHands
  • aider
  • Zed
  • Warp
  • Roo Code
  • Continue
  • Goose
  • OpenClaw
  • Crush
  • Qwen Code

Trust and governance

  1. Jurisdiction

    Built by a Canadian company, protected by Canadian privacy law.

    Compliant with PIPEDA and GDPR.

  2. Encryption

    Every saved conversation is encrypted.

    History is encrypted at rest and stays encrypted as it syncs across the web, desktop and phone.

  3. Ownership

    Your content is never sold and never shared with advertisers.

    Your prompts power Raven, nothing else. In DCode, agents ask before any command runs or any file is written.

Read the Privacy Policy

From the lab

Research, launches and engineering notes

Browse the newsroom
Three product marks side by side on a light ground: DCode's disc first and largest, then the Cursor logo and the GitHub Copilot logo.
Research

DCode vs Cursor vs Copilot: why we side with the agent

Autocomplete, the AI-first editor, or the full agent: a straight comparison of the three philosophies of AI-assisted coding, written by the people who built the third one.

A terminal window with two commands, npm install -g @dipoleml/dcode-cli and then dcode, beside the Raven Flash input price of $0.50 per 1M tokens and the DCode disc.
Developers

npm install and go: the DCode CLI and the real cost of the DipoleML API

The Centre Block on Parliament Hill in Ottawa, the Peace Tower under a deep blue sky, beside a checklist of five ticked lines marked two minutes.
Research

How we protect your data: a two-minute privacy check before you paste

The figure 1M in very large type, the context window of Raven Max and Raven Flash, beside 128K, their output limit, in smaller blue type, over a long strip of small page blocks, with the Raven app icon.
Research

A million tokens of room to think: what big context is actually for

We are an independent frontier AI lab, built in Toronto and bootstrapped from day one.

Our journeyCareersFor enterpriseFor the public sector

Put Raven to work

Three ways in, and one account opens all of them.

Try RavenFree to start on the web, with one account for web, desktop and phone.
Get DCodeVersion 2.5.0 for Windows and macOS, with a CLI for your terminal.
Build on the APIOpenAI and Anthropic formats. Prepaid credits never expire.

Founding Members lock in first-year annual pricing on Pro, Ultra and DCode. See pricing