diff --git a/README.md b/README.md
index 446e207f0..c70d3df33 100644
--- a/README.md
+++ b/README.md
@@ -308,7 +308,7 @@ Full reference: [Interaction and commands](docs/interaction.en.md).
 
 `/resume` · `/home` · `/agentview` · `/bg` · `⌸` open the same session manager: workspace rail, live state, filter, ★ pins. Also `/model` `/new` `/compact` `/export` `/btw` `/tree` `/fork` `/rewind` `/settings` `/setup` `/status` `/cost` `/jobs` `/skills` `/mcp` `/provider` `/auth` `/login` `/update`.
 
-`/model` shows models and reasoning effort on one page. DSH has provider tabs with **Recently used** first, moving the current model to the top; `Tab` / `Shift+Tab` select the next / previous provider. Codex and Claude open their model catalogs directly, without provider or recents tabs. `↑/↓` select a model, `←/→` adjust its effort, `Enter` applies the model and any explicit effort draft, and `Esc` cancels. Changing only the model preserves the backend's preference handling. Mouse users can click the available tabs, models, effort levels, and select/cancel hints; the wheel moves through models.
+`/model` shows models and reasoning effort on one page. DSH has provider tabs with **Recently used** first, moving the current model to the top; `Tab` / `Shift+Tab` select the next / previous provider. Codex and Claude open their model catalogs directly, without provider or recents tabs. `↑/↓` select a model, `←/→` adjust its effort, `Enter` applies the model and any explicit effort draft, and `Esc` cancels. Changing only the model preserves the backend's preference handling. Mouse users can click the available tabs, models, effort levels, and select/cancel hints; clicking an effort level immediately applies the focused model and that effort and closes the picker. The wheel moves through models.
 
 Short terminals keep the focused model and select/cancel actions visible, hiding descriptions and neighboring models first, then the effort row when only two rows remain.
 Shortcuts sit below the title and effort levels have their own section; DSH, Claude, and Codex use the terminal background while covering the text underneath.
diff --git a/README_ZH.md b/README_ZH.md
index e29c0f569..d28119737 100644
--- a/README_ZH.md
+++ b/README_ZH.md
@@ -262,7 +262,7 @@ OpenAI 条款约束。完整操作与当前边界：[Codex 后端](docs/codex-ba
 
 `/resume` · `/home` · `/agentview` · `/bg` · `⌸` 打开同一个会话管理界面：工作区栏、实时状态、筛选、★ 固定。另有 `/model` `/new` `/compact` `/export` `/btw` `/tree` `/fork` `/rewind` `/settings` `/setup` `/status` `/cost` `/jobs` `/skills` `/mcp` `/provider` `/auth` `/login` `/update`。
 
-`/model` 同页选择模型与推理等级。DSH 顶部显示提供商标签，**最近使用**排在第一位，打开时将当前模型去重置顶；`Tab` / `Shift+Tab` 切换下一个 / 上一个提供商。Codex 与 Claude 直接展示模型目录，不显示提供商或最近使用标签。`↑/↓` 选择模型，`←/→` 调整推理等级，`Enter` 应用模型及显式调整的推理草稿，`Esc` 取消；只换模型时沿用后端既有偏好处理。鼠标可点击可用的标签、模型、档位及选择/取消提示，滚轮切换模型。
+`/model` 同页选择模型与推理等级。DSH 顶部显示提供商标签，**最近使用**排在第一位，打开时将当前模型去重置顶；`Tab` / `Shift+Tab` 切换下一个 / 上一个提供商。Codex 与 Claude 直接展示模型目录，不显示提供商或最近使用标签。`↑/↓` 选择模型，`←/→` 调整推理等级，`Enter` 应用模型及显式调整的推理草稿，`Esc` 取消；只换模型时沿用后端既有偏好处理。鼠标可点击可用的标签、模型、档位及选择/取消提示；点击推理档位会立即应用当前聚焦模型与该强度，并关闭选择器。滚轮切换模型。
 
 短终端优先保留聚焦模型与选择/取消入口，先隐藏描述和邻项，仅剩两行时再隐藏推理行。
 操作提示位于标题下，推理档位独立成区；DSH、Claude、Codex 均沿用终端背景，并遮住底层文字。
diff --git a/docs/user-guide.en.md b/docs/user-guide.en.md
index c1b133753..428eee99b 100644
--- a/docs/user-guide.en.md
+++ b/docs/user-guide.en.md
@@ -481,7 +481,7 @@ Full-screen view of the whole session timeline (doesn't pollute scrollback); key
 - `/model`: selector. **Switching = fork the session** (history kept, only routing changes, the old session stays in `/resume`);
   persisted to `~/.dsh-tui/model.json` and reused by the next launch (a `provider`/`model`
   pair in cordis.yml/profile is only the first-run deployment default).
-- DSH provider tabs start with **Recently used**, and `Tab` / `Shift+Tab` switch providers; Codex and Claude show their catalogs directly without these tabs. All backends use `↑/↓` to select a model, `←/→` to adjust its effort, `Enter` to apply both, and `Esc` to cancel the draft. Click the available tabs, models, effort levels, or select/cancel hints; the wheel moves model focus.
+- DSH provider tabs start with **Recently used**, and `Tab` / `Shift+Tab` switch providers; Codex and Claude show their catalogs directly without these tabs. All backends use `↑/↓` to select a model, `←/→` to adjust its effort, `Enter` to apply both, and `Esc` to cancel the draft. Click the available tabs, models, effort levels, or select/cancel hints; clicking an effort level immediately applies the focused model and that effort and closes the picker. The wheel moves model focus.
 - Switching is rejected mid-turn.
 - `/preset` options: `standard` (default full features), `ptc`, `minimal` (the kernel's Minimal preset: one persistent-shell tool only, no compaction, no plan mode — and therefore no compaction and no tool-result pruning, so a long session can hit the context limit and `/compact` plus questions are unavailable; Help and `/` completion mark the command, and entering the preset says so once),
   `cordis`, `liangshen` (Liangshen mode).
diff --git a/docs/user-guide.md b/docs/user-guide.md
index 76fecb026..d8702e5f7 100644
--- a/docs/user-guide.md
+++ b/docs/user-guide.md
@@ -443,7 +443,7 @@ dsh-TUI 不预装通用技能；`/skills` 浏览 DSH 发现的技能，可直调
 - `/model`：选择器。**切换 = fork 会话续聊**（历史保留、仅换路由，旧会话留在 `/resume`）；
   持久化 `~/.dsh-tui/model.json`，下次启动直接沿用它（cordis.yml/profile 中的
   `provider`/`model` 只是首次启动的部署默认值）。
-- DSH 的顶部标签以「最近使用」开头，再列各提供商，`Tab` / `Shift+Tab` 切换提供商；Codex 与 Claude 直接显示模型目录，不显示这些标签。所有后端均用 `↑/↓` 选模型、`←/→` 调推理等级、`Enter` 一并应用、`Esc` 取消草稿，支持点击可用的标签、模型、档位与选择/取消提示，以及滚轮移动模型焦点。
+- DSH 的顶部标签以「最近使用」开头，再列各提供商，`Tab` / `Shift+Tab` 切换提供商；Codex 与 Claude 直接显示模型目录，不显示这些标签。所有后端均用 `↑/↓` 选模型、`←/→` 调推理等级、`Enter` 一并应用、`Esc` 取消草稿，支持点击可用的标签、模型、档位与选择/取消提示，以及滚轮移动模型焦点。点击推理档位会立即应用当前聚焦模型与该强度，并关闭选择器。
 - 回合运行中切换会被拒绝。
 - `/preset` 可选：`standard`（默认全功能）、`ptc`、`minimal`（内核「极简模式」：只暴露一个持久 shell 工具，无 compaction、无计划模式；也因此没有压缩、工具结果不剪枝——长会话可能撞上下文上限，`/compact` 与问卷不可用，Help 与 `/` 补全会标注，进入该预设时提示一次）、
   `cordis`、`liangshen`（梁神模式）。
diff --git a/guide/dsh-tui-guide/user-guide.en.md b/guide/dsh-tui-guide/user-guide.en.md
index c1b133753..428eee99b 100644
--- a/guide/dsh-tui-guide/user-guide.en.md
+++ b/guide/dsh-tui-guide/user-guide.en.md
@@ -481,7 +481,7 @@ Full-screen view of the whole session timeline (doesn't pollute scrollback); key
 - `/model`: selector. **Switching = fork the session** (history kept, only routing changes, the old session stays in `/resume`);
   persisted to `~/.dsh-tui/model.json` and reused by the next launch (a `provider`/`model`
   pair in cordis.yml/profile is only the first-run deployment default).
-- DSH provider tabs start with **Recently used**, and `Tab` / `Shift+Tab` switch providers; Codex and Claude show their catalogs directly without these tabs. All backends use `↑/↓` to select a model, `←/→` to adjust its effort, `Enter` to apply both, and `Esc` to cancel the draft. Click the available tabs, models, effort levels, or select/cancel hints; the wheel moves model focus.
+- DSH provider tabs start with **Recently used**, and `Tab` / `Shift+Tab` switch providers; Codex and Claude show their catalogs directly without these tabs. All backends use `↑/↓` to select a model, `←/→` to adjust its effort, `Enter` to apply both, and `Esc` to cancel the draft. Click the available tabs, models, effort levels, or select/cancel hints; clicking an effort level immediately applies the focused model and that effort and closes the picker. The wheel moves model focus.
 - Switching is rejected mid-turn.
 - `/preset` options: `standard` (default full features), `ptc`, `minimal` (the kernel's Minimal preset: one persistent-shell tool only, no compaction, no plan mode — and therefore no compaction and no tool-result pruning, so a long session can hit the context limit and `/compact` plus questions are unavailable; Help and `/` completion mark the command, and entering the preset says so once),
   `cordis`, `liangshen` (Liangshen mode).
diff --git a/guide/dsh-tui-guide/user-guide.md b/guide/dsh-tui-guide/user-guide.md
index 76fecb026..d8702e5f7 100644
--- a/guide/dsh-tui-guide/user-guide.md
+++ b/guide/dsh-tui-guide/user-guide.md
@@ -443,7 +443,7 @@ dsh-TUI 不预装通用技能；`/skills` 浏览 DSH 发现的技能，可直调
 - `/model`：选择器。**切换 = fork 会话续聊**（历史保留、仅换路由，旧会话留在 `/resume`）；
   持久化 `~/.dsh-tui/model.json`，下次启动直接沿用它（cordis.yml/profile 中的
   `provider`/`model` 只是首次启动的部署默认值）。
-- DSH 的顶部标签以「最近使用」开头，再列各提供商，`Tab` / `Shift+Tab` 切换提供商；Codex 与 Claude 直接显示模型目录，不显示这些标签。所有后端均用 `↑/↓` 选模型、`←/→` 调推理等级、`Enter` 一并应用、`Esc` 取消草稿，支持点击可用的标签、模型、档位与选择/取消提示，以及滚轮移动模型焦点。
+- DSH 的顶部标签以「最近使用」开头，再列各提供商，`Tab` / `Shift+Tab` 切换提供商；Codex 与 Claude 直接显示模型目录，不显示这些标签。所有后端均用 `↑/↓` 选模型、`←/→` 调推理等级、`Enter` 一并应用、`Esc` 取消草稿，支持点击可用的标签、模型、档位与选择/取消提示，以及滚轮移动模型焦点。点击推理档位会立即应用当前聚焦模型与该强度，并关闭选择器。
 - 回合运行中切换会被拒绝。
 - `/preset` 可选：`standard`（默认全功能）、`ptc`、`minimal`（内核「极简模式」：只暴露一个持久 shell 工具，无 compaction、无计划模式；也因此没有压缩、工具结果不剪枝——长会话可能撞上下文上限，`/compact` 与问卷不可用，Help 与 `/` 补全会标注，进入该预设时提示一次）、
   `cordis`、`liangshen`（梁神模式）。
diff --git a/scripts/verify-model-picker-ui.tsx b/scripts/verify-model-picker-ui.tsx
index a37ab3dbf..0e5c0f5b6 100644
--- a/scripts/verify-model-picker-ui.tsx
+++ b/scripts/verify-model-picker-ui.tsx
@@ -1,7 +1,7 @@
 /**
  * /model through the real Chat/channel/rendering path, with a fake LLM catalog:
  * recents-first tabs, current-model promotion, forward/backward wrapping, independent model/effort drafts,
- * same-batch navigation/confirmation, cancellation, mouse picks and wheel,
+ * same-batch navigation/confirmation, cancellation, single-click effort confirmation and wheel,
  * focus windowing and resize in inline/fullscreen at 100 and 36 columns;
  * header shortcuts, readable effort colors and the original terminal background.
  * Claude/Codex use a flat catalog, focus the current model after loading,
@@ -329,10 +329,19 @@ async function scenario(fullscreen: boolean, columns: number): Promise<void> {
     if (fullscreen) {
       await click('Beta 01')
       await check('mouse focuses a model', () => focused('Beta 01') && inverse('OFF'))
+      assert.deepEqual(switches, ['alpha/a1'], 'clicking a model only updates its draft')
       await click('MAX')
-      await check('mouse selects effort draft', () => inverse('MAX'))
-      await click('选择')
-      await check('mouse applies model and effort', () => channel.model === 'b1' && channel.reasoningEffort === 'max' && hit('最近使用') === undefined)
+      await check('one effort click applies the focused model and effort and closes the picker', () => channel.model === 'b1' && channel.reasoningEffort === 'max' && hit('最近使用') === undefined)
+      assert.deepEqual(switches, ['alpha/a1', 'beta/b1'])
+      assert.deepEqual(effortPicks, ['high', 'max'])
+      assert.equal(JSON.parse(readFileSync(join(prefsDir, 'effort.json'), 'utf8')).effort, 'max')
+      await open('Beta 01', 'MAX')
+      await click('beta / Beta 01')
+      await check('clicking the current model keeps the picker open', () => focused('Beta 01') && inverse('MAX'))
+      await click('MAX')
+      await check('clicking the selected effort also confirms and closes', () => hit('最近使用') === undefined && switches.length === 3 && effortPicks.length === 3)
+      assert.deepEqual(switches, ['alpha/a1', 'beta/b1', 'beta/b1'])
+      assert.deepEqual(effortPicks, ['high', 'max', 'max'])
       stdin.write('/model')
       await check('mouse follow-up command', () => hit('/model') !== undefined)
       stdin.write('\r')
@@ -441,9 +450,9 @@ async function backendSurface(backendId: 'claude' | 'codex', fullscreen: boolean
       await click('Model 01')
       assert.ok(await settled(() => focused('Model 01') && panelUsesDefaultBackground(term)), `${label}: mouse selects a model`)
       await click('HIGH')
-      await click('选择')
-      assert.ok(await settled(() => !text().includes('推理强度') && picks.length === 2 && effortPicks.at(-1) === 'high'), `${label}: mouse applies model and reasoning`)
+      assert.ok(await settled(() => !text().includes('推理强度') && picks.length === 2 && effortPicks.at(-1) === 'high'), `${label}: one effort click applies model and reasoning and closes the picker`)
       assert.deepEqual(picks, ['m0', 'm1'])
+      assert.deepEqual(effortPicks, ['high'])
     } else {
       assert.deepEqual(effortPicks, [], `${label}: no reasoning changes without an explicit draft`)
     }
diff --git a/src/components/ModelPicker.tsx b/src/components/ModelPicker.tsx
index 40a75510e..bd36b025a 100644
--- a/src/components/ModelPicker.tsx
+++ b/src/components/ModelPicker.tsx
@@ -184,6 +184,7 @@ export function ModelPicker({
                   muted={false}
                   cursor={caretZone === 'effort'}
                   onPick={onEffort}
+                  onConfirm={onConfirm}
                 />
               </>
             )}
@@ -197,7 +198,7 @@ export function ModelPicker({
 }
 
 /** Keep the active cell visible when a provider or effort strip exceeds its width. */
-function PickerTabs({ labels, focusIndex, width, muted = true, cursor = false, onPick }: {
+function PickerTabs({ labels, focusIndex, width, muted = true, cursor = false, onPick, onConfirm }: {
   labels: readonly string[]
   focusIndex: number
   width: number
@@ -205,6 +206,8 @@ function PickerTabs({ labels, focusIndex, width, muted = true, cursor = false, o
   /** This strip owns the native caret — its region was the last one used. */
   cursor?: boolean
   onPick(index: number): void
+  /** Confirm a clicked label; overflow arrows only navigate. */
+  onConfirm?(): void
 }): React.ReactNode {
   const singleLines = labels.map(label => label.replace(/[\r\n]+/g, ' '))
   const allFit = singleLines.reduce((sum, label) => sum + stringWidth(label) + 2, 0)
@@ -225,7 +228,10 @@ function PickerTabs({ labels, focusIndex, width, muted = true, cursor = false, o
           isFocused={start + index === focusIndex}
           muted={muted}
           cursor={cursor}
-          onPick={() => onPick(start + index)}
+          onPick={() => {
+            onPick(start + index)
+            onConfirm?.()
+          }}
         />
       ))}
       {end < cells.length ? <Box onClick={event => { event.stopImmediatePropagation(); onPick(end) }}><Text dimColor>›</Text></Box> : null}