fix(bedrock): resolve context limits for every vendor prefix, not just anthropic (#12921)

Boarded with 13 sibling PRs into one worktree off release/v3.8.51 and validated as a set: 132 focused tests pass across all 15 test files in the batch, typecheck:core is clean, check-changelog-integrity reports no lost base bullets, and check-file-size is green. Your PR merged without conflict against its siblings.

Thank you — the write-up made this reviewable: measuring the behaviour on the release tip and showing the before/after table meant the defect could be confirmed rather than taken on faith.

changelog.d/fixes/12921-bedrock-vendor-context-limits.md

@@ -0,0 +1 @@
+- **fix(bedrock):** model import now resolves context limits for every vendor prefix instead of only `anthropic.*`, so `global.openai.gpt-5.6-*` no longer imports with a null `inputTokenLimit` and gets rejected pre-flight at the 200k default ([#12921](https://github.com/diegosouzapw/OmniRoute/pull/12921)).

open-sse/config/bedrock.ts

@@ -90,13 +90,19 @@ export function getBedrockKnownModelLimits(modelId: string): {
   if (!trimmed) return null;
 
   const unqualified = trimmed.includes("/") ? trimmed.slice(trimmed.indexOf("/") + 1) : trimmed;
-  const withoutProfilePrefix = unqualified.replace(/^(?:eu|us|global)\./i, "");
-  const withoutProviderPrefix = withoutProfilePrefix.replace(/^anthropic\./i, "");
-  const spec =
-    getModelSpec(trimmed) ||
-    getModelSpec(unqualified) ||
-    getModelSpec(withoutProfilePrefix) ||
-    getModelSpec(withoutProviderPrefix);
+  // A Bedrock id is "<vendor>.<model>" optionally behind a cross-region profile
+  // prefix: "global.openai.gpt-5.6-sol", "us.anthropic.claude-...". The model
+  // name itself contains dots ("gpt-5.6-sol"), so peel at most those two leading
+  // qualifiers and keep the first candidate a spec knows. Peeling only
+  // "anthropic." left every other vendor (openai, meta, amazon, ...) without a
+  // context window, and the caller then fell back to a 200k default (#12915).
+  const segments = unqualified.split(".");
+  const spec = [trimmed, unqualified, segments.slice(1).join("."), segments.slice(2).join(".")]
+    .filter((candidate) => candidate.length > 0)
+    .reduce<ReturnType<typeof getModelSpec>>(
+      (found, candidate) => found || getModelSpec(candidate),
+      undefined
+    );
 
   if (!spec?.contextWindow && !spec?.maxOutputTokens) return null;
   return {

tests/unit/bedrock-vendor-context-limits-12915.test.ts

@@ -0,0 +1,57 @@
+import { describe, it } from "node:test";
+import assert from "node:assert/strict";
+import { discoverBedrockNativeModels } from "../../open-sse/services/bedrock.ts";
+
+// ─── #12915 — every Bedrock vendor prefix must resolve a context window ──────
+// Bedrock ids are "<vendor>.<model>", optionally behind a cross-region profile
+// prefix ("global.openai.gpt-5.6-sol"). The known-limits lookup used to peel
+// only "anthropic.", so imported openai.* models carried no inputTokenLimit and
+// the pre-flight context check fell back to a 200k default — rejecting 1M-context
+// models locally, before the request ever reached AWS.
+
+function bedrockFetcher(): (url: string, init: RequestInit) => Promise<Response> {
+  return async (url: string) => {
+    const body = url.includes("/inference-profiles")
+      ? { inferenceProfileSummaries: [] }
+      : {
+          modelSummaries: [
+            {
+              modelId: "global.openai.gpt-5.6-sol",
+              modelName: "GPT-5.6 Sol",
+              providerName: "OpenAI",
+              responseStreamingSupported: true,
+            },
+            {
+              modelId: "global.anthropic.claude-opus-4-6-v1",
+              modelName: "Claude Opus 4.6",
+              providerName: "Anthropic",
+              responseStreamingSupported: true,
+            },
+          ],
+        };
+    return new Response(JSON.stringify(body), {
+      status: 200,
+      headers: { "Content-Type": "application/json" },
+    });
+  };
+}
+
+describe("Bedrock model discovery (#12915)", () => {
+  it("carries a context window for openai.* models, not just anthropic.*", async () => {
+    const { models } = await discoverBedrockNativeModels({
+      apiKey: "test-key",
+      providerSpecificData: { region: "eu-west-1" },
+      fetcher: bedrockFetcher(),
+    });
+
+    const openai = models.find((m) => m.id === "global.openai.gpt-5.6-sol");
+    const anthropic = models.find((m) => m.id === "global.anthropic.claude-opus-4-6-v1");
+
+    // 1_050_000 and 1_000_000 differ, so a lookup that silently answered with the
+    // anthropic model's limit would not pass either assertion.
+    assert.equal(openai?.inputTokenLimit, 1_050_000);
+    assert.equal(openai?.outputTokenLimit, 128_000);
+    // The anthropic path must keep working unchanged.
+    assert.equal(anthropic?.inputTokenLimit, 1_000_000);
+  });
+});