Raven Model Family · Flagship Reasoning
Raven Max
Deep intelligence for high-stakes work
Raven Max is the flagship reasoning model in the Raven family, built for complex engineering, long-horizon planning, and decisions where errors are costly. It plans across multi-step dependencies, calls tools with precision, and streams extended reasoning before executing actions.
Specifications
| API model ID | raven-max |
| Modalities | Text + Vision |
| Context window | 1M tokens |
| Output limit | 128K tokens |
| Tool use | Native function calling |
| Input price | $1.25 / 1M tokens ($0.13 cached) |
| Output price | undefined |
| Release date | October 1, 2026 |
| Reasoning controls | reasoning_effort low or high, default when omitted, with an inspectable reasoning stream |
Prices are per 1M tokens in USD. See the pricing page for plans and credit packs.
Key workloads
Codebase Refactoring · Complex Architecture · Formal Research · Hard Bug Triage
Part of the family
Raven Flash pairs the same foundation with sub-second velocity for production speed.
Meet Raven FlashEmpirical evaluation
Benchmark results, as published.
Evaluated with maximum reasoning effort enabled across standard industry benchmarks. Source: Dipole ML Technical Report (October 2026).
| Benchmark | Raven Max | Rank | Best peer |
|---|---|---|---|
| Terminal Bench 4.0 | 42.8 | #1 | Fable 5 · 42.4 |
| Agent's Last Exam | 39.5 | #1 | Fable 5 · 31.6 |
| AutomationBench | 48.3 | #2 | Fable 5 · 48.6 |
| GDPVal-AA | 1771 | #1 | Fable 5 · 1760 |
| DeepSWE v1.1 | 68.7 | #4 | GPT-5.6 Sol · 70.2 |
| Intelligence Index | 47 | #2 | Fable 5 · 50 |
| HLE w/ Tools | 62.8 | #2 | Fable 5 · 63.2 |
| AA-Briefcase v1.1 | 1522 | #3 | Kimi K3 · 1547 |
Read the full evaluation in the Raven Max performance report, or see See the whole family compared.
Put Raven Maxto work.
Available in the Raven app, DCode, and through the OpenAI-compatible DipoleML API.