From 2d1a281d12ba002a2e8469f950446bd7860ca566 Mon Sep 17 00:00:00 2001
From: Koosha Paridehpour <42529354+KooshaPari@users.noreply.github.com>
Date: Mon, 14 Sep 2026 20:08:31 -0700
Subject: [PATCH] fix(providers): add alwaysReasons metadata for reka-flash-3
 (#13531)

`reka-flash-3` is marked `supportsReasoning` + `alwaysReasons` (with its 8192 output cap). `DefaultExecutor.ensureThinkingBudget` treats always-on reasoning as an implicit opt-in, so a small caller `max_tokens` gets the 4096 floor instead of being spent on reasoning and returning an empty answer (#13198). `alwaysReasons` is documented on `RegistryModel`.

Maintainer addition: `tests/unit/reka-flash-3-always-reasons-13198.test.ts` asserts the floor for `reka-flash-3` without any reasoning settings (fails on the release tip) and that `reka-flash` keeps the caller value. The branch also carried the #13532 Compare-scroll commit, which landed first; its hunks are identical, so this squash only adds the reka change.

Validated in one consolidated batch of this series (37 PRs boarded together on `release/v3.8.51`): `typecheck:core`, `check:open-sse-typecheck` and `check:dashboard-typecheck` clean; ESLint clean on every changed file; file-size, complexity, cognitive-complexity, changelog-integrity, docs-counts, docs-sync and migration-numbering gates green (only the pre-existing `open-sse/utils/stream.ts` file-size red remains, inherited from the base); 3,743 focused `node:test` cases plus 34 vitest cases green.

Thanks @KooshaPari!
---
 .../config/providers/registry/reka/index.ts   |  2 +-
 open-sse/config/providers/shared.ts           |  7 ++++++
 open-sse/executors/default.ts                 |  3 ++-
 .../reka-flash-3-always-reasons-13198.test.ts | 22 +++++++++++++++++++
 4 files changed, 32 insertions(+), 2 deletions(-)
 create mode 100644 tests/unit/reka-flash-3-always-reasons-13198.test.ts

diff --git a/open-sse/config/providers/registry/reka/index.ts b/open-sse/config/providers/registry/reka/index.ts
index b394800a09..c933eeb20d 100644
--- a/open-sse/config/providers/registry/reka/index.ts
+++ b/open-sse/config/providers/registry/reka/index.ts
@@ -11,7 +11,7 @@ export const rekaProvider: RegistryEntry = {
   models: [
     // reka-flash-3 stays first so it remains the provider default (the free-tier
     // model in freeModelCatalog); reka-flash was added in #4621 as an extra option.
-    { id: "reka-flash-3", name: "Reka Flash 3" },
+    { id: "reka-flash-3", name: "Reka Flash 3", supportsReasoning: true, alwaysReasons: true, maxOutputTokens: 8192 },
     { id: "reka-flash", name: "Reka Flash" },
     { id: "reka-edge-2603", name: "Reka Edge 2603" },
   ],
diff --git a/open-sse/config/providers/shared.ts b/open-sse/config/providers/shared.ts
index 8b9b45bfec..250d631775 100644
--- a/open-sse/config/providers/shared.ts
+++ b/open-sse/config/providers/shared.ts
@@ -56,6 +56,13 @@ export interface RegistryModel {
   liveCatalogIds?: readonly string[];
   toolCalling?: boolean;
   supportsReasoning?: boolean;
+  /**
+   * Model reasons unconditionally (always-on reasoning). When true,
+   * ensureThinkingBudget treats it as implicit reasoning opt-in so a tiny
+   * caller max_tokens gets the 4096 floor even without explicit thinking
+   * settings (#13198).
+   */
+  alwaysReasons?: boolean;
   supportedThinkingEfforts?: readonly string[];
   supportsVision?: boolean;
   supportsAudio?: boolean;
diff --git a/open-sse/executors/default.ts b/open-sse/executors/default.ts
index 4fb164a303..c267904d9f 100644
--- a/open-sse/executors/default.ts
+++ b/open-sse/executors/default.ts
@@ -1076,7 +1076,8 @@ export class DefaultExecutor extends BaseExecutor {
     const reasoningEnabled =
       thinking?.type === "enabled" ||
       (typeof effort === "string" && effort !== "none" && effort !== "off") ||
-      effort === true;
+      effort === true ||
+      modelEntry.alwaysReasons === true;
     if (!reasoningEnabled) return body;
 
     const MIN_TOKENS = 4096;
diff --git a/tests/unit/reka-flash-3-always-reasons-13198.test.ts b/tests/unit/reka-flash-3-always-reasons-13198.test.ts
new file mode 100644
index 0000000000..c74bdf54a3
--- /dev/null
+++ b/tests/unit/reka-flash-3-always-reasons-13198.test.ts
@@ -0,0 +1,22 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import { DefaultExecutor } from "../../open-sse/executors/default.ts";
+
+// #13198: reka-flash-3 reasons on every request, so a tiny caller max_tokens is
+// spent on reasoning and the answer comes back empty unless the 4096 floor applies
+// even without an explicit reasoning_effort / thinking setting.
+
+test("#13198 reka-flash-3 gets the thinking-budget floor without explicit reasoning settings", () => {
+  const executor = new DefaultExecutor("reka");
+  const body = { model: "reka-flash-3", max_tokens: 256 } as Record<string, unknown>;
+  executor.ensureThinkingBudget(body, "reka-flash-3");
+  assert.equal(body.max_tokens, 4096);
+});
+
+test("#13198 non-reasoning reka models keep the caller's max_tokens", () => {
+  const executor = new DefaultExecutor("reka");
+  const body = { model: "reka-flash", max_tokens: 256 } as Record<string, unknown>;
+  executor.ensureThinkingBudget(body, "reka-flash");
+  assert.equal(body.max_tokens, 256);
+});