NeuralTrust | The leading security platform for generative AI

The cost_cap policy (“LLM Cost Cap”) is a stateless, per-request price guard. Unlike the LLM Budget — which tracks cumulative spend over a window — cost cap looks only at the list price of the requested model and rejects or downgrades it before the call is made. It runs at pre_request. Use it to stop expensive models from being used at all on a given consumer or gateway, independent of volume.

Settings

Setting	Type	Default	Notes
`max_input_cost_per_1k_tokens`	number	—	Global input-price ceiling (USD per 1k tokens).
`max_output_cost_per_1k_tokens`	number	—	Global output-price ceiling (USD per 1k tokens).
`per_model_overrides`	map	—	Per-model ceilings keyed by model slug / wildcard (most specific wins).
`behavior_on_violation`	enum	`reject`	`reject` or `downgrade`.
`downgrade_to`	string	—	Target model for `downgrade` (must be on the same provider).
`unknown_model`	enum	`reject`	Price not resolvable: `reject`, `pass_through`, or `assume_max`.
`custom_pricing`	map	—	Per-token USD rates by model pattern, consulted before the built-in table.

Each per_model_overrides entry and custom_pricing value carries { input, output } price fields.

{
  "slug": "cost_cap",
  "settings": {
    "max_input_cost_per_1k_tokens": 0.005,
    "max_output_cost_per_1k_tokens": 0.015,
    "behavior_on_violation": "downgrade",
    "downgrade_to": "gpt-4o-mini",
    "unknown_model": "reject"
  }
}

Cost cap vs LLM Budget

	`cost_cap`	`token_rate_limiter` (LLM Budget)
State	Stateless (per request)	Stateful (Redis counters)
Guards	Model list price	Cumulative tokens or USD over a window
Use for	”Never let anyone use a model over $X/1k"	"Cap this tenant to $50/day”

Pair them: a cost cap blocks premium models outright, while an LLM Budget bounds total spend on the models that remain allowed.

​Settings

​Cost cap vs LLM Budget

Settings

Cost cap vs LLM Budget