Add Anthropic platform: wire-served metadata via listing dialect (provider 2)

The 5th registry row: id "anthropic", ANTHROPIC_API_KEY > auth.json
"anthropic" scope, api.anthropic.com/v1 with KIGI_ANTHROPIC_BASE_URL
override, Messages dialect. Two new spec dimensions most future rows
reuse: ListingDialect (Anthropic's /v1/models wants x-api-key +
anthropic-version headers, ?limit=1000, and its own response shape) and
PlatformKeyHeader (Bearer vs x-api-key across listing/validation/
inference, with auth_scheme stamped onto entries).

The 2026 Anthropic listing serves real metadata: the adapter maps
max_input_tokens, per-level effort capabilities (low..max as the menu,
xhigh/max distinct), thinking/image flags — and enrichment fills only
genuine wire gaps (e2e pins wire-1M beating enrichment, and a zero
context filled to 200k).

Two review-confirmed defects fixed red-green:
- Output caps were dropped at three layers, so every sub-128K-output
  model (64k Haiku, legacy models) would 400 on EVERY request against
  the sampler's 128K max_tokens default. Wire max_tokens and enrichment
  limit.output now flow to entry.max_completion_tokens.
- An explicit wire effort-decline was indistinguishable from wire
  silence, letting enrichment inject effort menus pre-4.6 models reject
  (adaptive thinking 400). The adapter now emits a decline sentinel
  (support:false) that enrichment respects — proven end to end.

Also: the Messages client now sends anthropic-version (previously never
sent — real api.anthropic.com rejects such requests; pinned across all
three scheme/backend quadrants), key validation builds per-key-header
requests, missing listing data fails fast, empty-id ghosts drop with a
warning, kimi-import recognizes api.anthropic.com as built-in
automatically.
This commit is contained in:
ZacharyZhang-NY committed 2026-07-21 06:12:53 -04:00
1 parent 23e94939c0
commit b86722f508
7 files changed
+579 -27

No files matched your search

+9 -3
View File
@@ -19,9 +19,8 @@ pub struct EnrichmentModel {
/// Max context window in tokens (`limit.context`).
#[serde(default, skip_serializing_if = "is_zero")]
pub context: u64,
/// Max output tokens (`limit.output`). Not yet consumed by
/// [`enrich_wire_model`]; feeds `max_completion_tokens` when provider
/// cycles start mapping it.
/// Max output tokens (`limit.output`); fills a wire-unserved output cap
/// (Anthropic 400s when `max_tokens` exceeds the model's limit).
#[serde(default, skip_serializing_if = "is_zero")]
pub output: u64,
/// Model supports reasoning/thinking.
@@ -186,6 +185,9 @@ pub fn enrich_wire_model(wire: &mut crate::WireModel, meta: &EnrichmentModel) {
if wire.context_length == 0 && meta.context > 0 {
wire.context_length = meta.context;
}
if wire.max_output_tokens == 0 && meta.output > 0 {
wire.max_output_tokens = meta.output;
}
if meta.reasoning {
wire.supports_reasoning = true;
}
@@ -235,6 +237,7 @@ mod tests {
fn enrich_fills_gaps_and_never_overwrites_wire() {
let meta = EnrichmentModel {
context: 400_000,
output: 64_000,
reasoning: true,
efforts: vec!["low".into(), "high".into()],
image_in: true,
@@ -246,6 +249,7 @@ mod tests {
serde_json::from_value(serde_json::json!({ "id": "gpt-test" })).unwrap();
enrich_wire_model(&mut bare, &meta);
assert_eq!(bare.context_length, 400_000);
assert_eq!(bare.max_output_tokens, 64_000, "output cap filled");
assert!(bare.supports_reasoning);
assert!(bare.supports_image_in);
assert_eq!(bare.display_name.as_deref(), Some("GPT Test"));
@@ -258,12 +262,14 @@ mod tests {
let mut served: crate::WireModel = serde_json::from_value(serde_json::json!({
"id": "gpt-test",
"context_length": 123,
"max_output_tokens": 77,
"display_name": "Wire Name",
"think_efforts": { "support": true, "valid_efforts": ["max"] }
}))
.unwrap();
enrich_wire_model(&mut served, &meta);
assert_eq!(served.context_length, 123, "wire context wins");
assert_eq!(served.max_output_tokens, 77, "wire output cap wins");
assert_eq!(served.display_name.as_deref(), Some("Wire Name"));
assert_eq!(
served.think_efforts.unwrap().valid_efforts,