Reference

AI Model Reference

Each skill in the Playbook is mapped to a model maturity level — L1 through L5 — that describes the kind of AI collaboration the skill requires. This page maps those levels to specific models so you can choose the right tool for the work.

Capability levels reflect what a model can do in a sustained workflow, not just its benchmark scores. Free tier availability is noted where relevant — cross-vendor adversarial review is easier when it costs nothing.

L2

Reasoning

Works through multi-step problems with guidance. Can analyse, summarise, and draft when given clear direction at each step.

Gemini 2.0 FlashGoogle

Fast, capable, and free. A strong choice for light drafting tasks and adversarial review cycles where cost matters.

Free tierAdversarial reviewResearchDrafting
GPT-4o MiniOpenAI

Cost-effective with solid instruction-following. Well-suited for research synthesis and lighter review cycles.

ResearchAdversarial review
L3

Agentic

Executes structured workflows with human oversight. Handles drafting, structuring, and critique across a full session with minimal per-step instruction.

Claude SonnetAnthropic

The balanced choice for sustained creative and analytical work. Strong instruction-following, long context, and natural voice.

Free tierDraftingLong-formAdversarial review
GPT-4oOpenAI

Versatile and widely deployed. Effective across full workflow cycles; strong for adversarial review from a different vendor.

Free tierDraftingAdversarial reviewResearch
Gemini 1.5 ProGoogle

Exceptional long-context handling with a generous free tier. A strong choice for adversarial review of longer pieces.

Free tierAdversarial reviewLong-formResearch
DeepSeek V3DeepSeek

Highly capable with API credits available. An independent model family perspective — valuable for cross-vendor adversarial review.

Free tierAdversarial reviewResearchDrafting
Mistral LargeMistral AI

European-hosted with a free tier. Strong reasoning and a distinct model family — useful when data residency matters or for cross-vendor review.

Free tierAdversarial reviewDrafting
L4

Autonomous

Plans and executes complex multi-step tasks with minimal human intervention. Extended reasoning, self-correction, and sustained autonomy over long tasks.

Claude OpusAnthropic

Extended reasoning with high editorial judgment. Appropriate for complex, high-stakes content where depth and nuance matter most.

DraftingLong-formReasoning
o3OpenAI

Strong extended reasoning. Well-suited to adversarial review requiring structured critique and logical depth.

Adversarial reviewReasoning
DeepSeek R1DeepSeek

Open-weight reasoning model with chain-of-thought. A strong open-source alternative at the extended reasoning tier.

Free tierReasoningAdversarial review
Gemini UltraGoogle

Google's most capable model for complex, multi-step tasks requiring sustained reasoning and broad knowledge synthesis.

ReasoningResearchLong-form