diff --git a/lib/ruby_llm/providers/bedrock/chat.rb b/lib/ruby_llm/providers/bedrock/chat.rb
index 026226c30..f125fd4c0 100644
--- a/lib/ruby_llm/providers/bedrock/chat.rb
+++ b/lib/ruby_llm/providers/bedrock/chat.rb
@@ -60,7 +60,7 @@ def parse_completion_response(response)
             output_tokens: usage['outputTokens'],
             cached_tokens: usage['cacheReadInputTokens'],
             cache_creation_tokens: usage['cacheWriteInputTokens'],
-            thinking_tokens: usage['reasoningTokens'],
+            thinking_tokens: usage['reasoningTokens'] || usage.dig('outputTokensDetails', 'reasoningTokens'),
             model_id: data['modelId'],
             raw: response
           )
diff --git a/spec/ruby_llm/providers/bedrock/chat_spec.rb b/spec/ruby_llm/providers/bedrock/chat_spec.rb
index 10e1d594a..430b9e145 100644
--- a/spec/ruby_llm/providers/bedrock/chat_spec.rb
+++ b/spec/ruby_llm/providers/bedrock/chat_spec.rb
@@ -128,4 +128,55 @@ def render_payload(messages = [], **overrides)
       end
     end
   end
+
+  describe '.parse_completion_response' do
+    it 'extracts thinking_tokens from top-level reasoningTokens' do
+      response_body = {
+        'output' => { 'message' => { 'content' => [{ 'text' => 'Hi!' }] } },
+        'usage' => {
+          'inputTokens' => 100,
+          'outputTokens' => 50,
+          'reasoningTokens' => 1200
+        }
+      }
+      response = instance_double(Faraday::Response, body: response_body)
+      message = described_class.parse_completion_response(response)
+
+      expect(message.thinking_tokens).to eq(1200)
+    end
+
+    it 'extracts thinking_tokens from nested outputTokensDetails' do
+      response_body = {
+        'output' => { 'message' => { 'content' => [{ 'text' => 'Hi!' }] } },
+        'usage' => {
+          'inputTokens' => 100,
+          'outputTokens' => 50,
+          'outputTokensDetails' => { 'reasoningTokens' => 800 }
+        }
+      }
+      response = instance_double(Faraday::Response, body: response_body)
+      message = described_class.parse_completion_response(response)
+
+      expect(message.thinking_tokens).to eq(800)
+    end
+
+    it 'captures cache usage metrics' do
+      response_body = {
+        'output' => { 'message' => { 'content' => [{ 'text' => 'Hi!' }] } },
+        'usage' => {
+          'inputTokens' => 100,
+          'outputTokens' => 50,
+          'cacheReadInputTokens' => 30,
+          'cacheWriteInputTokens' => 10
+        }
+      }
+      response = instance_double(Faraday::Response, body: response_body)
+      message = described_class.parse_completion_response(response)
+
+      expect(message.input_tokens).to eq(100)
+      expect(message.output_tokens).to eq(50)
+      expect(message.cached_tokens).to eq(30)
+      expect(message.cache_creation_tokens).to eq(10)
+    end
+  end
 end