diff --git a/service/convert.go b/service/convert.go
index a0ae8cb4df97..136157464f48 100644
--- a/service/convert.go
+++ b/service/convert.go
@@ -58,6 +58,9 @@ func ClaudeToOpenAIRequest(claudeRequest dto.ClaudeRequest, info *relaycommon.Re
 			openAIRequest.Reasoning = reasoningJSON
 		}
 	} else {
+		if effort := claudeRequest.GetEfforts(); effort != "" {
+			openAIRequest.ReasoningEffort = effort
+		}
 		thinkingSuffix := "-thinking"
 		if strings.HasSuffix(info.OriginModelName, thinkingSuffix) &&
 			!strings.HasSuffix(openAIRequest.Model, thinkingSuffix) {
diff --git a/service/convert_effort_test.go b/service/convert_effort_test.go
new file mode 100644
index 000000000000..e91ca4926e52
--- /dev/null
+++ b/service/convert_effort_test.go
@@ -0,0 +1,83 @@
+package service
+
+import (
+	"testing"
+
+	"github.com/QuantumNous/new-api/common"
+	"github.com/QuantumNous/new-api/constant"
+	"github.com/QuantumNous/new-api/dto"
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// Regression test for #5922: reasoning effort from a Claude-format request
+// (output_config.effort) must be forwarded as reasoning_effort when converting
+// to an OpenAI-format upstream request.
+func TestClaudeToOpenAIRequestForwardsEffortAsReasoningEffort(t *testing.T) {
+	var claudeRequest dto.ClaudeRequest
+	require.NoError(t, common.UnmarshalJsonStr(`{
+		"model": "gpt-5.2",
+		"max_tokens": 64,
+		"output_config": {"effort": "high"},
+		"messages": [
+			{"role": "user", "content": "hi"}
+		]
+	}`, &claudeRequest))
+
+	info := &relaycommon.RelayInfo{
+		ChannelMeta: &relaycommon.ChannelMeta{ChannelType: constant.ChannelTypeOpenAI},
+	}
+	openAIRequest, err := ClaudeToOpenAIRequest(claudeRequest, info)
+	require.NoError(t, err)
+	assert.Equal(t, "high", openAIRequest.ReasoningEffort)
+
+	encoded, err := common.Marshal(openAIRequest)
+	require.NoError(t, err)
+	assert.Contains(t, string(encoded), `"reasoning_effort":"high"`)
+}
+
+func TestClaudeToOpenAIRequestWithoutEffortLeavesReasoningEffortEmpty(t *testing.T) {
+	var claudeRequest dto.ClaudeRequest
+	require.NoError(t, common.UnmarshalJsonStr(`{
+		"model": "gpt-5.2",
+		"max_tokens": 64,
+		"messages": [
+			{"role": "user", "content": "hi"}
+		]
+	}`, &claudeRequest))
+
+	info := &relaycommon.RelayInfo{
+		ChannelMeta: &relaycommon.ChannelMeta{ChannelType: constant.ChannelTypeOpenAI},
+	}
+	openAIRequest, err := ClaudeToOpenAIRequest(claudeRequest, info)
+	require.NoError(t, err)
+	assert.Empty(t, openAIRequest.ReasoningEffort)
+
+	encoded, err := common.Marshal(openAIRequest)
+	require.NoError(t, err)
+	assert.NotContains(t, string(encoded), "reasoning_effort")
+}
+
+// OpenRouter conversion keeps its existing effort mapping and must not gain
+// a duplicate reasoning_effort field from this fix.
+func TestClaudeToOpenAIRequestOpenRouterEffortMappingUnchanged(t *testing.T) {
+	var claudeRequest dto.ClaudeRequest
+	require.NoError(t, common.UnmarshalJsonStr(`{
+		"model": "anthropic/claude-sonnet-4",
+		"max_tokens": 64,
+		"output_config": {"effort": "high"},
+		"messages": [
+			{"role": "user", "content": "hi"}
+		]
+	}`, &claudeRequest))
+
+	info := &relaycommon.RelayInfo{
+		ChannelMeta: &relaycommon.ChannelMeta{ChannelType: constant.ChannelTypeOpenRouter},
+	}
+	openAIRequest, err := ClaudeToOpenAIRequest(claudeRequest, info)
+	require.NoError(t, err)
+	assert.Empty(t, openAIRequest.ReasoningEffort)
+	assert.Equal(t, `"high"`, string(openAIRequest.Verbosity))
+}