Struct ModelSpec 

pub struct ModelSpec {

    pub provider: String,
    pub model: String,
    pub api_identifier: String,
    pub max_output_tokens: usize,
    pub input_context: usize,
    pub generation: f32,
    pub tier: String,
    pub legacy: bool,
    pub retires: Option<String>,
    pub supports_structured_output: bool,
    pub input_token_price: Option<f64>,
    pub output_token_price: Option<f64>,
    pub beta_headers: Vec<BetaHeader>,
    pub source: ModelSource,
}

Specification for a single model: its identity, limits, tier, and any beta-header unlocks.

A ModelSpec is the central row of the registry. provider and tier cross-reference into a ProviderConfig (via ModelConfiguration::providers and ProviderConfig::tiers). max_output_tokens and input_context are the base limits; entries in beta_headers raise them when the corresponding HTTP header is sent. source is loader-populated and records which layer contributed the entry — never read from YAML.

Identifier normalization

The same underlying model is addressable through several identifier formats depending on how the API is reached:

• Canonical (Anthropic direct): claude-3-7-sonnet-20250219
• Bedrock with region prefix: us.anthropic.claude-3-7-sonnet-20250219-v1:0
• AWS-direct without region: anthropic.claude-3-haiku-20240307-v1:0
• Regional gateways: eu.anthropic.claude-3-opus-20240229-v2:1

All four resolve to the same ModelSpec: ModelRegistry::get_model_spec tries an exact match first, and on miss strips region/provider prefixes and version suffixes before retrying. See ADR-0011 for the design rationale.

Fields

provider: String

AI provider name (e.g., “claude”).

model: String

Human-readable model name (e.g., “Claude Opus 4”).

api_identifier: String

API identifier used for requests (e.g., “claude-3-opus-20240229”).

max_output_tokens: usize

Maximum number of tokens that can be generated in a single response.

input_context: usize

Maximum number of tokens that can be included in the input context.

generation: f32

Model generation number (e.g., 3.0, 3.5, 4.0).

tier: String

Performance tier (e.g., “fast”, “balanced”, “flagship”).

legacy: bool

Whether this is a legacy model that may be deprecated.

retires: Option<String>

Announced retirement date (YYYY-MM-DD), if the provider has published one.

Complements Self::legacy, which cannot distinguish “older but fine” from “stops working on a known date” (#1334). None covers both models with no announced retirement and deprecated models whose date is still TBD — absence means “no published date”, never “not retiring”.

A past date means the model is already retired and its identifier now 404s; such entries are retained only so Bedrock/AWS identifier normalization keeps resolving them (see ADR-0011).

Held as a plain String: nothing parses or compares it today, it is carried verbatim into omni-dev config models show.

supports_structured_output: bool

Whether this model supports structured JSON-schema output via the Anthropic Messages API output_config.format (and the equivalent on Bedrock).

Gates the direct-API / Bedrock schema path (#1119): only models flagged here advertise AiClientCapabilities::supports_response_schema, so a caller on an older model — which would 400 on output_config — transparently keeps the YAML fallback. Defaults to false, so unmarked and unknown models are never sent the field.

input_token_price: Option<f64>

Price per million input tokens in USD, if known.

Used to compute per-invocation cost for backends that report token usage (currently the direct Anthropic API — see [crate::claude::ai::compute_cost_usd]). None for unpriced models (e.g. the OpenAI/Gemini entries), which surface cost as unknown.

output_token_price: Option<f64>

Price per million output tokens in USD, if known. See Self::input_token_price.

beta_headers: Vec<BetaHeader>

Beta headers that unlock enhanced limits for this model.

source: ModelSource

Layer that contributed this entry. Populated by the loader; never read from YAML.

Trait Implementations

impl Clone for ModelSpec

fn clone(&self) -> ModelSpec

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ModelSpec

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for ModelSpec

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ModelSpec

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.