Cut your AI cost IN HALF (EASY)

Technique of using a frontier model for planning and a cheaper good-enough model for code execution. The creator's worked example cuts cost from $9.50 to $3.02 (about 68%).

GPGPT-5.5OpenAIAvailable in EU

Used as the cheaper coding model that implements Fable's spec via Codex. The creator says OpenAI's generous quotas make it effectively cheap.

Frontier, most expensive model ($10/M input, $50/M output) used for research, spec writing and PR review in the creator's workflow.

COCodexOpenAIAvailable in EU

OpenAI's coding agent. It has a CLI, so Claude Code can call it from a skill that delegates code writing to GPT-5.5.

CCClaude CodeAnthropicAvailable in EU

Can call Codex through its CLI. The creator built a skill where Fable plans, Codex GPT-5.5 writes and Claude verifies.

CUCursorSpaceX (Cursor / Anysphere)Available in EU

Third-party harness with an auto mode that routes tasks to appropriate models. Its sub-agents delegate work to Composer 2.5.

C2Composer 2.5CursorAvailable in EU

Cursor's homegrown, inexpensive coding model. Cursor sub-agents use it for easier tasks even when a frontier model is selected.

NDNot DiamondNot DiamondUnverified

Company dedicated to model routing for enterprises. It can save money and often improves quality. The creator discloses being a small investor.

CSClaude Sonnet 5AnthropicAvailable in EU

The creator walks through the model picker. Haiku 4.5 is fast and cheap, Sonnet 5 is cheaper and still strong, Opus 4.8 is the previous generation but capable, and Fable is the best and costliest. He advises choosing models and thinking effort per task.

G5GLM 5.2Zhipu AIUnverified

Cheap open-source model that Coinbase uses for most coding tasks, helping keep costs flat while token usage grows.