diff --git a/src/benchmarks/benchmark-config.test.ts b/src/benchmarks/benchmark-config.test.ts
index 4cba901a..5aa74440 100644
--- a/src/benchmarks/benchmark-config.test.ts
+++ b/src/benchmarks/benchmark-config.test.ts
@@ -38,13 +38,14 @@ describe("benchmark config", () => {
     assertLeft(result);
   });
 
-  it("requires reasoningEffort in model benchmark configs", () => {
+  it("accepts model benchmark configs without reasoningEffort", () => {
     const result = parseSchema(BenchmarkRunConfigSchema, {
       benchmarkId: "gpqa_diamond",
       model: "openai/gpt-5",
     });
 
-    assertLeft(result);
+    assertRight(result);
+    expect(result.right).not.toHaveProperty("reasoningEffort");
   });
 
   it("parses injected benchmark configs with opaque options", () => {
diff --git a/src/benchmarks/benchmark-config.ts b/src/benchmarks/benchmark-config.ts
index 24f309ea..668676f3 100644
--- a/src/benchmarks/benchmark-config.ts
+++ b/src/benchmarks/benchmark-config.ts
@@ -44,7 +44,7 @@ export type GeminiMediaResolution = ValueOf<typeof GEMINI_MEDIA_RESOLUTIONS>;
 export const InferenceOverrideSchema = z.object({
   temperature: z.number().optional(),
   maxTokens: z.number().optional(),
-  reasoningEffort: z.enum(REASONING_EFFORTS),
+  reasoningEffort: z.enum(REASONING_EFFORTS).optional(),
   costTier: z.enum(COST_TIERS).optional(),
   timeoutMs: z.number().optional(),
   sort: z.nativeEnum(ProviderSort).optional(),
diff --git a/src/benchmarks/search/core/request.test.ts b/src/benchmarks/search/core/request.test.ts
index da276ff8..60b349b5 100644
--- a/src/benchmarks/search/core/request.test.ts
+++ b/src/benchmarks/search/core/request.test.ts
@@ -196,6 +196,10 @@ describe("buildSearchRequestBody", () => {
     expect(body.temperature).toBe(0);
     expect(body.reasoning).toEqual({ effort: "high" });
   });
+  it("omits reasoning when no effort is requested", () => {
+    const body = buildSearchRequestBody({ ...BASE, lane: lane({}) });
+    expect(body).not.toHaveProperty("reasoning");
+  });
   it("threads the candidate models list and omits it by default", () => {
     const routed = buildSearchRequestBody({
       ...BASE,
diff --git a/src/benchmarks/search/core/request.ts b/src/benchmarks/search/core/request.ts
index c2cf6588..aec4730b 100644
--- a/src/benchmarks/search/core/request.ts
+++ b/src/benchmarks/search/core/request.ts
@@ -28,7 +28,7 @@ export interface SearchRequestOptions {
   readonly lane: SearchLaneConfig;
   readonly maxOutputTokens?: number;
   readonly temperature?: number;
-  readonly reasoningEffort: ReasoningEffort;
+  readonly reasoningEffort?: ReasoningEffort;
   readonly sort?: ProviderSort;
   readonly providerOrder?: readonly string[];
   readonly providerOnly?: readonly string[];
@@ -127,7 +127,10 @@ export function buildSearchRequestBody(
       models: opts.models === undefined ? undefined : [...opts.models],
       maxOutputTokens: opts.maxOutputTokens,
       temperature: opts.temperature,
-      reasoning: { effort: opts.reasoningEffort },
+      reasoning:
+        opts.reasoningEffort !== undefined
+          ? { effort: opts.reasoningEffort }
+          : undefined,
       provider:
         opts.sort !== undefined ||
         opts.providerOrder !== undefined ||
diff --git a/src/benchmarks/search/core/solver.ts b/src/benchmarks/search/core/solver.ts
index f9aae09a..2adfbd28 100644
--- a/src/benchmarks/search/core/solver.ts
+++ b/src/benchmarks/search/core/solver.ts
@@ -50,7 +50,7 @@ export interface SearchSolverOptions {
   readonly timeoutMs?: number;
   readonly maxOutputTokens?: number;
   readonly temperature?: number;
-  readonly reasoningEffort: ReasoningEffort;
+  readonly reasoningEffort?: ReasoningEffort;
   readonly endpointId?: string;
   readonly sort?: ProviderSort;
   readonly providerOrder?: readonly string[];
diff --git a/src/cli/index.test.ts b/src/cli/index.test.ts
index 326bff7b..5d20a922 100644
--- a/src/cli/index.test.ts
+++ b/src/cli/index.test.ts
@@ -12,6 +12,28 @@ describe("bench-harness CLI", () => {
     );
   });
 
+  it("leaves reasoning effort unset for --reasoning-effort default", () => {
+    const args = parseArgs([
+      "--benchmark",
+      "gpqa_diamond",
+      "--model",
+      "openrouter/jev",
+      "--reasoning-effort",
+      "default",
+    ]);
+    expect(args.reasoningEffort).toBeUndefined();
+    const config = buildBenchmarkConfig({
+      benchmarkId: "gpqa_diamond",
+      model: "openrouter/jev",
+      panelConfig: undefined,
+      artifactDir: undefined,
+      endpointId: undefined,
+      imageDetail: undefined,
+      reasoningEffort: args.reasoningEffort,
+    });
+    expect(config).not.toHaveProperty("reasoningEffort");
+  });
+
   it("rejects an invalid --reasoning-effort value", () => {
     expect(() => parseArgs(["--reasoning-effort", "invalid"])).toThrow(
       "--reasoning-effort must be one of"
diff --git a/src/cli/index.ts b/src/cli/index.ts
index 4c9aea8d..f04cec47 100644
--- a/src/cli/index.ts
+++ b/src/cli/index.ts
@@ -46,7 +46,7 @@ interface CliArgs {
   readonly resumeId?: string;
   readonly imageDetail?: ImageDetail;
   readonly costTier?: CostTier;
-  readonly reasoningEffort: ReasoningEffort;
+  readonly reasoningEffort: ReasoningEffort | undefined;
 }
 
 export function parseArgs(argv: readonly string[]): CliArgs {
@@ -190,7 +190,7 @@ function main(): Promise<void> {
         reasoningEffort: args.reasoningEffort,
       });
       process.stderr.write(
-        `Running ${args.benchmark}${args.model !== undefined ? ` on ${args.model}` : ""}${args.solverConfig !== undefined ? ` (solver-config=${args.solverConfig})` : ""}${artifactDir !== undefined ? ` (artifact-dir=${artifactDir})` : ""} (epochs=${epochs}, concurrency=${args.concurrency}, reasoning-effort=${args.reasoningEffort}${range !== undefined ? `, range=${range.start ?? 0}..${range.end ?? "end"}` : ""}, session=${sessionId})...\n`
+        `Running ${args.benchmark}${args.model !== undefined ? ` on ${args.model}` : ""}${args.solverConfig !== undefined ? ` (solver-config=${args.solverConfig})` : ""}${artifactDir !== undefined ? ` (artifact-dir=${artifactDir})` : ""} (epochs=${epochs}, concurrency=${args.concurrency}, reasoning-effort=${args.reasoningEffort ?? MODEL_DEFAULT_REASONING_EFFORT}${range !== undefined ? `, range=${range.start ?? 0}..${range.end ?? "end"}` : ""}, session=${sessionId})...\n`
       );
       const total = yield* promise(() =>
         resolveTotalEvaluations(args.benchmark, range, epochs)
@@ -304,13 +304,20 @@ function validateCostTier(raw: string | undefined): CostTier | undefined {
   return raw;
 }
 
-function validateReasoningEffort(raw: string | undefined): ReasoningEffort {
+const MODEL_DEFAULT_REASONING_EFFORT = "default";
+
+function validateReasoningEffort(
+  raw: string | undefined
+): ReasoningEffort | undefined {
   if (raw === undefined) {
     return DEFAULT_REASONING_EFFORT;
   }
+  if (raw === MODEL_DEFAULT_REASONING_EFFORT) {
+    return undefined;
+  }
   if (!isMember(raw, REASONING_EFFORTS)) {
     throw new Error(
-      `--reasoning-effort must be one of: ${REASONING_EFFORTS.join(", ")} (got "${raw}")`
+      `--reasoning-effort must be one of: ${[...REASONING_EFFORTS, MODEL_DEFAULT_REASONING_EFFORT].join(", ")} (got "${raw}")`
     );
   }
   return raw;
@@ -330,7 +337,7 @@ function buildSchemaValidatedConfig(opts: {
   panelConfig: unknown;
   imageDetail?: ImageDetail;
   costTier?: CostTier;
-  reasoningEffort: ReasoningEffort;
+  reasoningEffort: ReasoningEffort | undefined;
 }): BenchmarkRunConfig {
   const {
     benchmarkId,
@@ -372,6 +379,7 @@ function buildSchemaValidatedConfig(opts: {
     : undefined;
   if (
     optionsSchema !== undefined &&
+    reasoningEffort !== undefined &&
     Object.hasOwn(optionsSchema.shape, "agentReasoningEffort") &&
     !(
       typeof panelConfig === "object" &&
@@ -410,7 +418,7 @@ export function buildBenchmarkConfig(opts: {
   endpointId: string | undefined;
   imageDetail: ImageDetail | undefined;
   costTier?: CostTier;
-  reasoningEffort: ReasoningEffort;
+  reasoningEffort: ReasoningEffort | undefined;
 }): BenchmarkRunConfig {
   const {
     benchmarkId,
diff --git a/src/harness/model.ts b/src/harness/model.ts
index 0e9d4b5f..0f85c282 100644
--- a/src/harness/model.ts
+++ b/src/harness/model.ts
@@ -20,7 +20,7 @@ export interface GenerateConfig {
   readonly maxTokens?: number;
   readonly endpointId?: string;
   readonly tools?: readonly ToolDefinition[];
-  readonly reasoningEffort: ReasoningEffort;
+  readonly reasoningEffort?: ReasoningEffort;
   readonly costTier?: CostTier;
   readonly timeoutMs?: number;
   readonly sort?: ProviderSort;
diff --git a/src/providers/responses-model.test.ts b/src/providers/responses-model.test.ts
index e3dd04e2..ec840278 100644
--- a/src/providers/responses-model.test.ts
+++ b/src/providers/responses-model.test.ts
@@ -313,6 +313,7 @@ describe("responses-model", () => {
       },
     ]);
     expect(sentOptions?.extraBody).toBeUndefined();
+    expect(sentBody).not.toHaveProperty("reasoning");
     expect(exit.value.outputItems).toEqual([
       {
         type: "function_call",
diff --git a/src/providers/responses-model.ts b/src/providers/responses-model.ts
index b558d63d..81293e59 100644
--- a/src/providers/responses-model.ts
+++ b/src/providers/responses-model.ts
@@ -185,8 +185,11 @@ export function generate(
           ? [...genConfig.tools]
           : undefined,
     }),
-    reasoning: { effort: genConfig.reasoningEffort },
     ...definedValues({
+      reasoning:
+        genConfig.reasoningEffort !== undefined
+          ? { effort: genConfig.reasoningEffort }
+          : undefined,
       provider: sendProvider ? providerPreferences : undefined,
     }),
     ...definedValues({