diff --git a/web/app/components/evaluation/__tests__/index.spec.tsx b/web/app/components/evaluation/__tests__/index.spec.tsx
index 4f8db5a32d..4a8d409d9a 100644
--- a/web/app/components/evaluation/__tests__/index.spec.tsx
+++ b/web/app/components/evaluation/__tests__/index.spec.tsx
@@ -128,7 +128,7 @@ const renderWithQueryClient = (ui: ReactNode) => {
 
 describe('Evaluation', () => {
   beforeEach(() => {
-    useEvaluationStore.setState({ resources: {} })
+    useEvaluationStore.setState({ resources: {}, initialResources: {} })
     vi.clearAllMocks()
     mockUseEvaluationConfig.mockReturnValue({
       data: null,
@@ -251,6 +251,37 @@ describe('Evaluation', () => {
     })
   })
 
+  it('should reset unsaved non-pipeline config changes to the hydrated config', () => {
+    mockUseEvaluationConfig.mockReturnValue({
+      data: {
+        evaluation_model: 'gpt-4o-mini',
+        evaluation_model_provider: 'openai',
+        default_metrics: [],
+        customized_metrics: null,
+        judgment_config: null,
+      },
+    })
+
+    renderWithQueryClient(<Evaluation resourceType="apps" resourceId="app-reset" />)
+
+    const resetButton = screen.getByRole('button', { name: 'common.operation.reset' })
+    expect(resetButton).toBeDisabled()
+
+    fireEvent.click(screen.getByRole('button', { name: 'evaluation.metrics.add' }))
+    fireEvent.change(screen.getByPlaceholderText('evaluation.metrics.searchNodeOrMetrics'), {
+      target: { value: 'faith' },
+    })
+    fireEvent.click(screen.getByTestId('evaluation-metric-node-faithfulness-node-faithfulness'))
+
+    expect(useEvaluationStore.getState().resources['apps:app-reset']!.metrics).toHaveLength(1)
+    expect(resetButton).toBeEnabled()
+
+    fireEvent.click(resetButton)
+
+    expect(useEvaluationStore.getState().resources['apps:app-reset']!.metrics).toHaveLength(0)
+    expect(resetButton).toBeDisabled()
+  })
+
   it('should hide the value row for empty operators', () => {
     const resourceType = 'apps'
     const resourceId = 'app-2'
diff --git a/web/app/components/evaluation/__tests__/store.spec.ts b/web/app/components/evaluation/__tests__/store.spec.ts
index 150f285b52..648eee68d8 100644
--- a/web/app/components/evaluation/__tests__/store.spec.ts
+++ b/web/app/components/evaluation/__tests__/store.spec.ts
@@ -10,7 +10,7 @@ import { buildEvaluationConfigPayload, buildEvaluationRunRequest } from '../stor
 
 describe('evaluation store', () => {
   beforeEach(() => {
-    useEvaluationStore.setState({ resources: {} })
+    useEvaluationStore.setState({ resources: {}, initialResources: {} })
   })
 
   it('should configure a custom metric mapping to a valid state', () => {
diff --git a/web/app/components/evaluation/components/batch-test-panel/index.tsx b/web/app/components/evaluation/components/batch-test-panel/index.tsx
index ce3ace9ef5..8b4d9ca98a 100644
--- a/web/app/components/evaluation/components/batch-test-panel/index.tsx
+++ b/web/app/components/evaluation/components/batch-test-panel/index.tsx
@@ -1,13 +1,9 @@
 'use client'
 
 import type { BatchTestTab, EvaluationResourceProps } from '../../types'
-import { Button } from '@langgenius/dify-ui/button'
 import { cn } from '@langgenius/dify-ui/cn'
-import { toast } from '@langgenius/dify-ui/toast'
 import { useTranslation } from 'react-i18next'
-import { useSaveEvaluationConfigMutation } from '@/service/use-evaluation'
 import { isEvaluationRunnable, useEvaluationResource, useEvaluationStore } from '../../store'
-import { buildEvaluationConfigPayload } from '../../store-utils'
 import { TAB_CLASS_NAME } from '../../utils'
 import HistoryTab from './history-tab'
 import InputFieldsTab from './input-fields-tab'
@@ -19,63 +15,21 @@ const BatchTestPanel = ({
   resourceId,
 }: EvaluationResourceProps) => {
   const { t } = useTranslation('evaluation')
-  const { t: tCommon } = useTranslation('common')
   const tabLabels: Record<BatchTestTab, string> = {
     'input-fields': t('batch.tabs.input-fields'),
     'history': t('batch.tabs.history'),
   }
   const resource = useEvaluationResource(resourceType, resourceId)
   const setBatchTab = useEvaluationStore(state => state.setBatchTab)
-  const saveConfigMutation = useSaveEvaluationConfigMutation()
   const isRunnable = isEvaluationRunnable(resource)
   const isPanelReady = !!resource.judgeModelId && resource.metrics.length > 0
 
-  const handleSave = () => {
-    if (!isRunnable) {
-      toast.warning(t('batch.validation'))
-      return
-    }
-
-    const body = buildEvaluationConfigPayload(resource, resourceType)
-
-    if (!body) {
-      toast.warning(t('batch.validation'))
-      return
-    }
-
-    saveConfigMutation.mutate({
-      params: {
-        targetType: resourceType,
-        targetId: resourceId,
-      },
-      body,
-    }, {
-      onSuccess: () => {
-        toast.success(tCommon('api.saved'))
-      },
-      onError: () => {
-        toast.error(t('config.saveFailed'))
-      },
-    })
-  }
-
   return (
     <div className="flex h-full min-h-0 flex-col bg-background-default">
       <div className="px-6 py-4">
-        <div className="flex items-start justify-between gap-3">
-          <div className="min-w-0">
-            <div className="system-xl-semibold text-text-primary">{t('batch.title')}</div>
-            <div className="mt-1 system-sm-regular text-text-tertiary">{t('batch.description')}</div>
-          </div>
-          <Button
-            className="shrink-0"
-            variant="primary"
-            disabled={!isRunnable}
-            loading={saveConfigMutation.isPending}
-            onClick={handleSave}
-          >
-            {tCommon('operation.save')}
-          </Button>
+        <div className="min-w-0">
+          <div className="system-xl-semibold text-text-primary">{t('batch.title')}</div>
+          <div className="mt-1 system-sm-regular text-text-tertiary">{t('batch.description')}</div>
         </div>
         <div className="mt-4 rounded-xl border border-divider-subtle bg-components-card-bg p-3">
           <div className="flex items-start gap-3">
diff --git a/web/app/components/evaluation/components/config-actions.tsx b/web/app/components/evaluation/components/config-actions.tsx
new file mode 100644
index 0000000000..8c1e769045
--- /dev/null
+++ b/web/app/components/evaluation/components/config-actions.tsx
@@ -0,0 +1,80 @@
+'use client'
+
+import type { EvaluationResourceProps } from '../types'
+import { Button } from '@langgenius/dify-ui/button'
+import { toast } from '@langgenius/dify-ui/toast'
+import { useTranslation } from 'react-i18next'
+import { useSaveEvaluationConfigMutation } from '@/service/use-evaluation'
+import {
+  isEvaluationRunnable,
+  useEvaluationResource,
+  useEvaluationStore,
+  useIsEvaluationConfigDirty,
+} from '../store'
+import { buildEvaluationConfigPayload } from '../store-utils'
+
+const EvaluationConfigActions = ({
+  resourceType,
+  resourceId,
+}: EvaluationResourceProps) => {
+  const { t } = useTranslation('evaluation')
+  const { t: tCommon } = useTranslation('common')
+  const resource = useEvaluationResource(resourceType, resourceId)
+  const isDirty = useIsEvaluationConfigDirty(resourceType, resourceId)
+  const resetResourceConfig = useEvaluationStore(state => state.resetResourceConfig)
+  const markResourceConfigSaved = useEvaluationStore(state => state.markResourceConfigSaved)
+  const saveConfigMutation = useSaveEvaluationConfigMutation()
+  const isRunnable = isEvaluationRunnable(resource)
+
+  const handleSave = () => {
+    if (!isRunnable) {
+      toast.warning(t('batch.validation'))
+      return
+    }
+
+    const body = buildEvaluationConfigPayload(resource, resourceType)
+
+    if (!body) {
+      toast.warning(t('batch.validation'))
+      return
+    }
+
+    saveConfigMutation.mutate({
+      params: {
+        targetType: resourceType,
+        targetId: resourceId,
+      },
+      body,
+    }, {
+      onSuccess: () => {
+        markResourceConfigSaved(resourceType, resourceId)
+        toast.success(tCommon('api.saved'))
+      },
+      onError: () => {
+        toast.error(t('config.saveFailed'))
+      },
+    })
+  }
+
+  return (
+    <div className="flex shrink-0 items-center gap-2">
+      <Button
+        variant="secondary"
+        disabled={!isDirty || saveConfigMutation.isPending}
+        onClick={() => resetResourceConfig(resourceType, resourceId)}
+      >
+        {tCommon('operation.reset')}
+      </Button>
+      <Button
+        variant="primary"
+        disabled={!isRunnable}
+        loading={saveConfigMutation.isPending}
+        onClick={handleSave}
+      >
+        {tCommon('operation.save')}
+      </Button>
+    </div>
+  )
+}
+
+export default EvaluationConfigActions
diff --git a/web/app/components/evaluation/components/layout/non-pipeline-evaluation.tsx b/web/app/components/evaluation/components/layout/non-pipeline-evaluation.tsx
index 5d47a754ff..5c6acf289d 100644
--- a/web/app/components/evaluation/components/layout/non-pipeline-evaluation.tsx
+++ b/web/app/components/evaluation/components/layout/non-pipeline-evaluation.tsx
@@ -5,6 +5,7 @@ import { useTranslation } from 'react-i18next'
 import { useDocLink } from '@/context/i18n'
 import BatchTestPanel from '../batch-test-panel'
 import ConditionsSection from '../conditions-section'
+import EvaluationConfigActions from '../config-actions'
 import JudgeModelSelector from '../judge-model-selector'
 import MetricSection from '../metric-section'
 import SectionHeader, { InlineSectionHeader } from '../section-header'
@@ -38,6 +39,7 @@ const NonPipelineEvaluation = ({
               </>
             )}
             descriptionClassName="max-w-[700px]"
+            action={<EvaluationConfigActions resourceType={resourceType} resourceId={resourceId} />}
           />
           <section className="max-w-[700px] py-4">
             <InlineSectionHeader title={t('judgeModel.title')} tooltip={t('judgeModel.description')} />
diff --git a/web/app/components/evaluation/components/layout/pipeline-evaluation.tsx b/web/app/components/evaluation/components/layout/pipeline-evaluation.tsx
index 9fb6c70a90..102360f526 100644
--- a/web/app/components/evaluation/components/layout/pipeline-evaluation.tsx
+++ b/web/app/components/evaluation/components/layout/pipeline-evaluation.tsx
@@ -6,6 +6,7 @@ import { useTranslation } from 'react-i18next'
 import { useDocLink } from '@/context/i18n'
 import { useEvaluationStore } from '../../store'
 import HistoryTab from '../batch-test-panel/history-tab'
+import EvaluationConfigActions from '../config-actions'
 import JudgeModelSelector from '../judge-model-selector'
 import PipelineBatchActions from '../pipeline/pipeline-batch-actions'
 import PipelineMetricsSection from '../pipeline/pipeline-metrics-section'
@@ -45,6 +46,7 @@ const PipelineEvaluation = ({
                 </a>
               </>
             )}
+            action={<EvaluationConfigActions resourceType={resourceType} resourceId={resourceId} />}
           />
         </div>
 
diff --git a/web/app/components/evaluation/store.ts b/web/app/components/evaluation/store.ts
index 4e2df1bb00..9f94e594b0 100644
--- a/web/app/components/evaluation/store.ts
+++ b/web/app/components/evaluation/store.ts
@@ -4,6 +4,7 @@ import type {
   EvaluationResourceType,
 } from './types'
 import type { EvaluationConfig, NodeInfo } from '@/types/evaluation'
+import { isEqual } from 'es-toolkit/predicate'
 import { create } from 'zustand'
 import { getEvaluationMockConfig } from './mock'
 import {
@@ -28,8 +29,11 @@ import { buildConditionMetricOptions } from './utils'
 
 type EvaluationStore = {
   resources: Record<string, EvaluationResourceState>
+  initialResources: Record<string, EvaluationResourceState>
   ensureResource: (resourceType: EvaluationResourceType, resourceId: string) => void
   hydrateResource: (resourceType: EvaluationResourceType, resourceId: string, config: EvaluationConfig) => void
+  resetResourceConfig: (resourceType: EvaluationResourceType, resourceId: string) => void
+  markResourceConfigSaved: (resourceType: EvaluationResourceType, resourceId: string) => void
   setJudgeModel: (resourceType: EvaluationResourceType, resourceId: string, judgeModelId: string) => void
   addBuiltinMetric: (resourceType: EvaluationResourceType, resourceId: string, optionId: string, nodeInfoList?: NodeInfo[]) => void
   updateMetricThreshold: (resourceType: EvaluationResourceType, resourceId: string, metricId: string, threshold: number) => void
@@ -88,8 +92,68 @@ type EvaluationStore = {
 
 const initialResourceCache: Record<string, EvaluationResourceState> = {}
 
+const cloneEvaluationResourceState = (resource: EvaluationResourceState): EvaluationResourceState => ({
+  ...resource,
+  metrics: resource.metrics.map(metric => ({
+    ...metric,
+    nodeInfoList: metric.nodeInfoList?.map(nodeInfo => ({ ...nodeInfo })),
+    customConfig: metric.customConfig
+      ? {
+          ...metric.customConfig,
+          mappings: metric.customConfig.mappings.map(mapping => ({ ...mapping })),
+          outputs: metric.customConfig.outputs.map(output => ({ ...output })),
+        }
+      : undefined,
+  })),
+  judgmentConfig: {
+    ...resource.judgmentConfig,
+    conditions: resource.judgmentConfig.conditions.map(condition => ({ ...condition })),
+  },
+  batchRecords: resource.batchRecords.map(record => ({ ...record })),
+})
+
+const preserveBatchState = (
+  configState: EvaluationResourceState,
+  currentResource: EvaluationResourceState | undefined,
+  resourceType: EvaluationResourceType,
+): EvaluationResourceState => {
+  const initialState = buildInitialState(resourceType)
+
+  return {
+    ...cloneEvaluationResourceState(configState),
+    activeBatchTab: currentResource?.activeBatchTab ?? initialState.activeBatchTab,
+    uploadedFileId: currentResource?.uploadedFileId ?? initialState.uploadedFileId,
+    uploadedFileName: currentResource?.uploadedFileName ?? initialState.uploadedFileName,
+    selectedRunId: currentResource?.selectedRunId ?? initialState.selectedRunId,
+    batchRecords: currentResource?.batchRecords.map(record => ({ ...record })) ?? initialState.batchRecords,
+  }
+}
+
+const createConfigSnapshot = (
+  resourceType: EvaluationResourceType,
+  resource: EvaluationResourceState,
+): EvaluationResourceState => {
+  const initialState = buildInitialState(resourceType)
+
+  return {
+    ...cloneEvaluationResourceState(resource),
+    activeBatchTab: initialState.activeBatchTab,
+    uploadedFileId: initialState.uploadedFileId,
+    uploadedFileName: initialState.uploadedFileName,
+    selectedRunId: initialState.selectedRunId,
+    batchRecords: initialState.batchRecords,
+  }
+}
+
+const pickConfigComparableState = (resource: EvaluationResourceState) => ({
+  judgeModelId: resource.judgeModelId,
+  metrics: resource.metrics,
+  judgmentConfig: resource.judgmentConfig,
+})
+
 export const useEvaluationStore = create<EvaluationStore>((set, get) => ({
   resources: {},
+  initialResources: {},
   ensureResource: (resourceType, resourceId) => {
     const resourceKey = buildResourceKey(resourceType, resourceId)
     if (get().resources[resourceKey])
@@ -103,17 +167,42 @@ export const useEvaluationStore = create<EvaluationStore>((set, get) => ({
     }))
   },
   hydrateResource: (resourceType, resourceId, config) => {
+    const resourceKey = buildResourceKey(resourceType, resourceId)
+    const configState = buildStateFromEvaluationConfig(resourceType, config)
+
     set(state => ({
       resources: {
         ...state.resources,
-        [buildResourceKey(resourceType, resourceId)]: {
-          ...buildStateFromEvaluationConfig(resourceType, config),
-          activeBatchTab: state.resources[buildResourceKey(resourceType, resourceId)]?.activeBatchTab ?? 'input-fields',
-          uploadedFileId: state.resources[buildResourceKey(resourceType, resourceId)]?.uploadedFileId ?? null,
-          uploadedFileName: state.resources[buildResourceKey(resourceType, resourceId)]?.uploadedFileName ?? null,
-          selectedRunId: state.resources[buildResourceKey(resourceType, resourceId)]?.selectedRunId ?? null,
-          batchRecords: state.resources[buildResourceKey(resourceType, resourceId)]?.batchRecords ?? [],
-        },
+        [resourceKey]: preserveBatchState(configState, state.resources[resourceKey], resourceType),
+      },
+      initialResources: {
+        ...state.initialResources,
+        [resourceKey]: createConfigSnapshot(resourceType, configState),
+      },
+    }))
+  },
+  resetResourceConfig: (resourceType, resourceId) => {
+    const resourceKey = buildResourceKey(resourceType, resourceId)
+
+    set(state => ({
+      resources: {
+        ...state.resources,
+        [resourceKey]: preserveBatchState(
+          state.initialResources[resourceKey] ?? buildInitialState(resourceType),
+          state.resources[resourceKey],
+          resourceType,
+        ),
+      },
+    }))
+  },
+  markResourceConfigSaved: (resourceType, resourceId) => {
+    const resourceKey = buildResourceKey(resourceType, resourceId)
+    const resource = get().resources[resourceKey] ?? buildInitialState(resourceType)
+
+    set(state => ({
+      initialResources: {
+        ...state.initialResources,
+        [resourceKey]: createConfigSnapshot(resourceType, resource),
       },
     }))
   },
@@ -436,6 +525,20 @@ export const useEvaluationResource = (resourceType: EvaluationResourceType, reso
   return useEvaluationStore(state => state.resources[resourceKey] ?? (initialResourceCache[resourceKey] ??= buildInitialState(resourceType)))
 }
 
+export const useIsEvaluationConfigDirty = (resourceType: EvaluationResourceType, resourceId: string) => {
+  const resourceKey = buildResourceKey(resourceType, resourceId)
+
+  return useEvaluationStore((state) => {
+    const resource = state.resources[resourceKey] ?? (initialResourceCache[resourceKey] ??= buildInitialState(resourceType))
+    const initialResource = state.initialResources[resourceKey] ?? buildInitialState(resourceType)
+
+    return !isEqual(
+      pickConfigComparableState(resource),
+      pickConfigComparableState(initialResource),
+    )
+  })
+}
+
 export const getAllowedOperators = (
   metrics: EvaluationResourceState['metrics'],
   variableSelector: [string, string] | null,