Returns the full model catalog, the curated picker lists (Kai agent, Kai Code cloud, Kai Code OpenCode, transcription), and the per-runtime default model ids. Internal fields and markup metadata are not exposed; customer-facing cost figures are pre-computed.
Ok
Full catalog (every entry, no filtering).
Effective raw rate per 1M tokens. Always populated for catalog entries (pure-alias tiers resolve theirs from the backing model).
Prompt-cache write rate for the default 5-minute TTL (Anthropic: 1.25× input).
Prompt-cache write rate for the 1-hour TTL (Anthropic ttl: '1h': 2× input).
getModelCost fills it in as 2× input for Anthropic models that don't
set it; every other provider has no 1h tier and prices a 1h bucket at the
5-minute write rate.
Anthropic models that reject temperature (claude-opus-4-8, claude-haiku-5-5).
Anthropic models whose thinking can't be turned off: thinking: {type:'disabled'} is a 400 at every effort level (Fable 5 / 5.1,
Opus 5.5). Low effort is the only way to make them think less — see
claudeReasoningOptions.
Per-runtime UI-picker visibility.
kaiAgent → Kai default support agent dropdown
kaiCodeCloud → Kai Code cloud runtime model picker
kaiCodeOpenCode → Kai Code desktop OpenCode runtime model picker
Models excluded from every picker but still in the catalog are
either subagent-only (e.g. Kimi K2.5) or special-purpose (transcription,
Codex CLI) — they appear elsewhere via capabilities.transcription
or via the static staticModels list of CLI-only runtimes.
Customer-facing average cost (USD) for a typical Kai support answer. Pre-multiplied by the Kai markup so the Frontend can render the figure without ever seeing the multiplier.
What Gleap bills per 1M tokens for this model, by where it runs
(effective raw rate × that usage type's markup). Pickers show these;
costPerMillionTokens stays the raw provider rate.
Kai support answers (kai markup).
Custom agents, Kai Resolve and its code investigations, Kai Copilot (default markup).
Kai Code cloud sessions (gleapcode markup).
Curated picker lists (replaces FE iterate-and-filter logic).
STT models (transcription-capable).
Value persisted on the project / session. Either an alias or a canonical id.
Kai Code OpenCode (desktop) runtime model picker.
Value persisted on the project / session. Either an alias or a canonical id.
Kai Code cloud runtime model picker.
Value persisted on the project / session. Either an alias or a canonical id.
Default Kai support agent dropdown.
Value persisted on the project / session. Either an alias or a canonical id.
Per-runtime default modelId.