fix: pass through max tokens for non-mantle Bedrock models

src/assets/__tests__/__snapshots__/assets.snapshot.test.ts.snap

@@ -6199,7 +6199,7 @@ from strands.models.bedrock import BedrockModel
 
 def load_model() -> BedrockModel:
     """Get Bedrock model client using IAM credentials."""
-    return BedrockModel(model_id="{{#if modelId}}{{modelId}}{{else}}global.anthropic.claude-sonnet-4-5-20250929-v1:0{{/if}}")
+    return BedrockModel(model_id="{{#if modelId}}{{modelId}}{{else}}global.anthropic.claude-sonnet-4-5-20250929-v1:0{{/if}}"{{#if modelMaxTokens}}, max_tokens={{modelMaxTokens}}{{/if}})
 {{/if}}
 {{/if}}
 {{#if (eq modelProvider "Anthropic")}}

src/assets/python/http/strands/base/model/load.py

@@ -65,7 +65,7 @@ def load_model():
 
 def load_model() -> BedrockModel:
     """Get Bedrock model client using IAM credentials."""
-    return BedrockModel(model_id="{{#if modelId}}{{modelId}}{{else}}global.anthropic.claude-sonnet-4-5-20250929-v1:0{{/if}}")
+    return BedrockModel(model_id="{{#if modelId}}{{modelId}}{{else}}global.anthropic.claude-sonnet-4-5-20250929-v1:0{{/if}}"{{#if modelMaxTokens}}, max_tokens={{modelMaxTokens}}{{/if}})
 {{/if}}
 {{/if}}
 {{#if (eq modelProvider "Anthropic")}}

src/cli/commands/export/__tests__/harness-mapper.test.ts

@@ -858,6 +858,36 @@ describe('model ID propagation to renderConfig', () => {
   });
 });
 
+// ============================================================================
+// model maxTokens propagation
+// ============================================================================
+
+describe('model maxTokens propagation to renderConfig', () => {
+  it('propagates maxTokens for an ordinary Converse Bedrock model', () => {
+    const ctx = baseContext({
+      model: { provider: 'bedrock', modelId: 'anthropic.claude-3', maxTokens: 4096 },
+    });
+    const { renderConfig } = mapHarnessToExportConfig(ctx, 'CodeZip');
+    expect(renderConfig.modelMaxTokens).toBe(4096);
+    expect(renderConfig.bedrockMantle).toBeUndefined();
+  });
+
+  it('leaves modelMaxTokens undefined when the spec has no maxTokens', () => {
+    const ctx = baseContext({ model: { provider: 'bedrock', modelId: 'anthropic.claude-3' } });
+    const { renderConfig } = mapHarnessToExportConfig(ctx, 'CodeZip');
+    expect(renderConfig.modelMaxTokens).toBeUndefined();
+  });
+
+  it('propagates maxTokens for a Bedrock Mantle model', () => {
+    const ctx = baseContext({
+      model: { provider: 'bedrock', modelId: 'openai.gpt-oss-120b', apiFormat: 'chat_completions', maxTokens: 2048 },
+    });
+    const { renderConfig } = mapHarnessToExportConfig(ctx, 'CodeZip');
+    expect(renderConfig.bedrockMantle).toBe(true);
+    expect(renderConfig.modelMaxTokens).toBe(2048);
+  });
+});
+
 // ============================================================================
 // resolveIdentityProvider
 // ============================================================================

src/cli/commands/export/harness-mapper.ts

@@ -273,6 +273,9 @@
     hasExecutionLimits,
     isExportHarness: true,
     modelId: spec.model.modelId,
+    // Max output tokens, threaded into load.py for both ordinary Converse Bedrock models
+    // (BedrockModel max_tokens) and Mantle models (max_output_tokens/max_completion_tokens).
+    modelMaxTokens: spec.model.maxTokens,
     // LiteLLM-only model config (apiBase + additionalParams), threaded into load.py. The model
     // schema is a flat object, so apiBase/additionalParams are always typed-present — only the
     // provider check + a truthiness check are needed.
@@ -556,7 +559,6 @@
     mantleProprietary: isProprietaryOpenAiModel(spec.model.modelId),
     modelTemperature: spec.model.temperature,
     modelTopP: spec.model.topP,
-    modelMaxTokens: spec.model.maxTokens,
   };
 }
 
@@ -1204,7 +1206,7 @@
 }
 
 function fnmatch(pattern: string, str: string): boolean {
   const re = new RegExp(
     '^' +
       pattern
         .replace(/[.+^${}()|[\]\\]/g, '\\$&')

src/cli/templates/types.ts

@@ -168,7 +168,8 @@ export interface AgentRenderConfig {
   modelTemperature?: number;
   /** Model nucleus-sampling top_p (export path, Mantle): merged into the client params when set. */
   modelTopP?: number;
-  /** Model max output tokens (export path, Mantle): mapped to max_output_tokens / max_completion_tokens. */
+  /** Model max output tokens (export path): BedrockModel max_tokens for Converse models;
+   *  max_output_tokens / max_completion_tokens for Mantle models. */
   modelMaxTokens?: number;
 
   /** True when generating from a harness export (suppresses placeholder tools) */