Try Raven

Models

Raven Max and Raven Flash, the Raven model family

Raven Max and Raven Flash are the two frontier language models built in Canada by Dipole ML Inc. Both read text and images, call tools natively, show their reasoning, and accept up to 1M tokens of context. Raven Max is the flagship for deep reasoning; Raven Flash is the fast one for everyday and high-volume work.

Open the playgroundView pricing
  • 42.8First on Terminal Bench 4.0Raven Max
  • 39.5First on Agent's Last ExamRaven Max
  • 49.2First on AutomationBenchRaven Flash
  • 1771First on GDPVal-AARaven Max

Which Raven model should I use?

Raven Max · Flagship reasoning

Precision for high-stakes decisions

Built for complex engineering, long-horizon planning, and decisions where errors are costly. Raven Max plans across multi-step dependencies, calls tools with precision, and streams extended reasoning before it acts.

Key workloads

  • Codebase Refactoring
  • Complex Architecture
  • Formal Research
  • Hard Bug Triage

Benchmark wins

  • #1 on Terminal Bench 4.0 (42.8)
  • #1 on Agent's Last Exam (39.5)
  • #1 on GDPVal-AA (1771)

Each model is ranked among the models it is compared with in the benchmarks below.

$1.25 input, $0.13 cached input, $6.25 output, per 1M tokens

About Raven MaxTest in the playground

Raven Flash · High-velocity frontier

Frontier intelligence at production speed

Engineered for low latency, responsive autonomous agents, and high-frequency code execution without compromising accuracy. Raven Flash streams inspectable reasoning and completions in real time.

Key workloads

  • Autonomous Coding in DCode
  • Interactive Chat
  • CI/CD Pipelines
  • High-Volume Classification

Benchmark wins

  • #1 on Terminal Bench 4.0 (34.1)
  • #1 on AutomationBench (49.2)
  • Tied #1 on Intelligence Index (43)

Each model is ranked among the models it is compared with in the benchmarks below.

$0.50 input, $0.05 cached input, $2.50 output, per 1M tokens

About Raven FlashTest in the playground

Empirical Evaluation

How do Raven models score on public benchmarks?

Evaluated with maximum reasoning effort enabled across standard industry benchmarks. Raven models are natively optimized for terminal execution, autonomous tool use, and multi-step reasoning.

Raven Max, flagship evaluations

Raven Max
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
Fable 548.6
Raven Max48.3
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
DeepSWE v1.1
ModelScore
GPT-5.6 Sol70.2
Fable 569.7
Gemini 3.8 Flash69.4
Raven Max68.7
Muse Spark 1.368.0
Intelligence Index
ModelScore
Fable 550
Raven Max47
Muse Spark 1.345
GPT-5.6 Sol44
Gemini 3.8 Flash41
HLE w/ Tools
ModelScore
Fable 563.2
Raven Max62.8
Kimi K359.8
GPT-5.6 Sol58.5
Gemini 3.8 Flash53.8
AA-Briefcase v1.1
ModelScore
Kimi K31547
Fable 51539
Raven Max1522
GPT-5.6 Sol1520
Muse Spark 1.31475

Evaluated with maximum reasoning effort enabled. Source: Dipole ML Inc. Technical Report (October 2026).

Raven Flash, velocity evaluations

Raven Flash
34.1
GPT-5.6 Terra
24.6
Claude Opus 4.8
21.7
DeepSeek-V4-Vision
18.2
Gemini 3.7 Flash
17.5
Terminal Bench 4.0
ModelScore
Raven Flash34.1
GPT-5.6 Terra24.6
Claude Opus 4.821.7
DeepSeek-V4-Vision18.2
Gemini 3.7 Flash17.5
Agent's Last Exam
ModelScore
Claude Opus 4.827.0
Raven Flash26.6
GPT-5.6 Terra25.1
Gemini 3.7 Flash22.4
DeepSeek-V4-Vision21.8
AutomationBench
ModelScore
Raven Flash49.2
Claude Opus 4.841.0
DeepSeek-V4-Vision38.8
Gemini 3.7 Flash38.6
GPT-5.6 Terra37.2
GDPVal-AA
ModelScore
Claude Opus 4.81731
Raven Flash1651
GPT-5.6 Terra1571
Gemini 3.7 Flash1527
DeepSeek-V4-Vision1510
DeepSWE v1.1
ModelScore
GPT-5.6 Terra69.6
Raven Flash65.4
Gemini 3.7 Flash65.3
DeepSeek-V4-Vision59.3
Claude Opus 4.858.0
Intelligence Index
ModelScore
Raven Flash43
Claude Opus 4.843
Muse Glimmer43
Gemini 3.7 Flash41
GPT-5.6 Terra37
HLE w/ Tools
ModelScore
Claude Opus 4.857.9
Raven Flash55.8
GPT-5.6 Terra52.1
DeepSeek-V4-Vision51.0
Gemini 3.7 Flash49.8
AA-Briefcase v1.1
ModelScore
Claude Opus 4.81480
Raven Flash1462
GPT-5.6 Terra1385
DeepSeek-V4-Vision1266
Gemini 3.7 Flash1132

Evaluated with maximum reasoning effort enabled. Source: Dipole ML Inc. Technical Report (October 2026).

How does Raven work?

One continuous architecture behind every Raven response.

Thinking you can inspect

Raven models stream their chain of thought in collapsible, real-time blocks. Developers and security auditors can check the reasoning before any code is written or any outside tool is called.

Reads images alongside text

Raven models read architecture diagrams, PDF schematics, UI wireframes and terminal error logs natively alongside code, without separate vision adapters.

Calls tools and reports what it did

Tuned for multi-step agentic work: shell commands, file edits, git workflows and outside API requests, with a structured receipt for every action.

What do Raven models cost?

Text + Vision · Native function calling

ModelInput / 1MCached input / 1MOutput / 1M
Raven Max"Raven Max"$1.25$0.13$6.25
Raven Flash"Raven Flash"$0.50$0.05$2.50

Prices are per 1M tokens in USD. The API supports both /v1/chat/completions (OpenAI SDKs) and /v1/messages (Anthropic SDKs): switch your base URL to api.dipoleml.com.

Work out your monthly cost on the API pageView pricing

Where can I use Raven models?

Raven app

Chat, images and thinking blocks, on the web and in the apps.

About the app

DCode

Agentic coding with Flash by default and Max for the deep work.

Explore DCode

DipoleML API

OpenAI and Anthropic compatible, streaming, pay as you go with prepaid credits.

Read the docs

Playground

The real models, free, right in your browser.

Open the playground

Try Raven in your browser

The playground runs the real models with streaming responses, inspectable thinking blocks and native vision. Sign in free to try it.

Open the playgroundView pricing