Related to #16755.

With no modelRoles.vision or @default set, an image sent to a text-only model is described by the first image-capable chat model in catalog order (resolveVisionModel in packages/coding-agent/src/utils/image-vision-fallback.ts). For an Anthropic-only user, that model is claude-fable-5.

• Expensive: $10 input / $50 output per million tokens. Sonnet 4.6 is $3/$15 and Haiku 4.5 is $1/$5. Fable 5 is tied for the highest Anthropic price, though not one of the most expensive catalog entries (231 of 5,948 have input at or above $10/Mtok).
• Outdated: claude-fable-5-1 (revision 5.1.0) is in the same catalog at the same input and output price.

Each attached image is one call to this model.

A cost-based selection (least expensive priced image-capable model) is in 5914a04 (5914a04), which is not yet on main or in v18.13.0.