docs(benchmarks): focus leaderboard on aggregate results

README.md

@@ -27,6 +27,15 @@
   <img src="https://github.com/RLinf/misc/raw/main/pic/rpent_framework.png" alt="RPent framework"/>
 </div>
 
+## Benchmark Results
+
+Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see [Benchmark Results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) for suite results and model configurations.
+
+Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results and memory-batch explanation](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html#libero-pro-astra-memory), including the separately frozen Long and Spatial/Object/Goal memory batches.
+
+[![RPent success-rate comparisons across four benchmarks](https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html)
+
+
 ## Who Should Consider Using RPent?
 
 RPent is built for four kinds of users:
@@ -38,6 +47,7 @@ RPent is built for four kinds of users:
 
 ## What's NEW!
 
+- [2026/09] 🔥 Added an interactive leaderboard and consolidated benchmark results for LIBERO, LIBERO-PRO, RoboCasa365, and RoboTwin, with model comparisons and suite-level results. Explore [Benchmark Results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html).
 - [2026/09] 🔥 RPent supports Franka single-arm and dual-arm real-robot extensions. Doc: [Franka](https://rpent.readthedocs.io/en/latest/rst_source/usage/franka.html) · [Dual Franka](https://rpent.readthedocs.io/en/latest/rst_source/usage/dual_franka.html).
 - [2026/08] 🔥 RPent supports RoboCasa with RLDX-1 as manipulation model. See the [RoboCasa setup and Target50 guide](https://rpent.readthedocs.io/en/latest/rst_source/usage/robocasa.html).
 - [2026/08] 🔥 RPent supports the non-reasoning mode, which reduces average execution time by ~40%.

README.zh-CN.md

@@ -27,6 +27,15 @@
   <img src="https://github.com/RLinf/misc/raw/main/pic/rpent_framework.png" alt="RPent framework"/>
 </div>
 
+## 基准测试结果
+
+对比 RPent 与参考方法在 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。排名仅限图中方法及评测范围；套件成绩和模型配置见[基准测试结果](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)。
+
+Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总与 memory 批次说明](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html#libero-pro-astra-memory)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。
+
+[![RPent 四项基准成功率对比](https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)
+
+
 ## 适用用户
 
 RPent 面向以下四类用户：
@@ -38,6 +47,7 @@ RPent 面向以下四类用户：
 
 ## 最新动态
 
+- [2026/09] 🔥 新增交互式排行榜和基准测试结果汇总，覆盖 LIBERO、LIBERO-PRO、RoboCasa365 与 RoboTwin，提供模型对比及套件汇总。查看[基准测试结果](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)。
 - [2026/08] 🔥 支持 RoboCasa，使用 RLDX-1 作为操作模型。参见 [RoboCasa 安装与 Target50 指南](https://rpent.readthedocs.io/zh-cn/latest/rst_source/usage/robocasa.html)。
 - [2026/08] 🔥 新增非推理（non-reasoning）模式，平均执行时间降低约 40%。
 - [2026/08] 🔥 支持 LIBERO 探索模式。文档：[LIBERO 探索模式](https://rpent.readthedocs.io/zh-cn/latest/rst_source/usage/libero.html#memory)。

docs/source-en/index.rst

@@ -74,6 +74,7 @@ Welcome to RPent
    Overview <rst_source/overview>
    Installation <rst_source/installation>
    Quick Start <rst_source/quickstart>
+   Benchmark Results <rst_source/benchmarks>
 
 .. toctree::
    :maxdepth: 2

docs/source-en/rst_source/benchmarks.rst

@@ -0,0 +1,24 @@
+:html_theme.sidebar_secondary.remove:
+
+.. _benchmark-results:
+.. _benchmark-leaderboard:
+.. _leaderboard:
+
+RPent Leaderboard
+=================
+
+.. raw:: html
+
+   <link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/docs.css">
+   <script defer src="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/table-sort.js"></script>
+   <script defer src="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard.js"></script>
+   <script defer src="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/embed.js"></script>
+   <div id="rpent-interactive-leaderboard" data-language="en"
+        data-results-url="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/results.json">
+     <div class="rpent-static-leaderboard">
+       <section data-benchmark="libero-pro"><h2>LIBERO-PRO</h2><table><thead><tr><th>Method / model</th><th>Success rate</th></tr></thead><tbody><tr><td>Codex / GPT-6 Astra / low / reasoning</td><td>92.63%</td></tr><tr><td>Claude Code / Opus-4.7 / max.reasoning</td><td>82.4%</td></tr><tr><td>RPent Flash Mode</td><td>72.63%</td></tr><tr><td>Codex / GPT-5.5 / xhigh / reasoning</td><td>72.1%</td></tr><tr><td>ASPIRE</td><td>61.36%</td></tr><tr><td>π_RLinf</td><td>50.0%</td></tr><tr><td>π0.5</td><td>11.0%</td></tr><tr><td>AtomVLA</td><td>6.3%</td></tr><tr><td>X-VLA</td><td>3.8%</td></tr><tr><td>MolmoAct</td><td>1.5%</td></tr><tr><td>π0</td><td>0.3%</td></tr></tbody></table><p>GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).</p><h3>All methods &amp; reported scores</h3><div class="table-wrap" tabindex="0"><table><thead><tr><th>Method / model</th><th>Overall</th><th>Spatial Task</th><th>Spatial Swap</th><th>Object Task</th><th>Object Swap</th><th>Goal Task</th><th>Goal Swap</th><th>Long Task</th><th>Long Swap</th></tr></thead><tbody><tr><td>GPT-6 Astra</td><td>92.63%</td><td>100%</td><td>98%</td><td>100%</td><td>99%</td><td>88%</td><td>99%</td><td>85%</td><td>72%</td></tr><tr><td>Opus-4.7 / max.reasoning</td><td>82.4%</td><td>—</td><td>—</td><td>—</td><td>—</td><td>—</td><td>—</td><td>—</td><td>—</td></tr><tr><td>RPent Flash Mode / Molmo2-8B</td><td>72.63%</td><td>79.00%</td><td>70.00%</td><td>86.00%</td><td>93.00%</td><td>74.00%</td><td>65.00%</td><td>60.00%</td><td>54.00%</td></tr><tr><td>GPT-5.5</td><td>72.1%</td><td>81.0%</td><td>69.0%</td><td>94.0%</td><td>91.0%</td><td>75.0%</td><td>66.0%</td><td>52.0%</td><td>49.0%</td></tr><tr><td>ASPIRE</td><td>61.36%</td><td>60.0%</td><td>51.0%</td><td>95.0%</td><td>98.0%</td><td>45.0%</td><td>81.0%</td><td>38.3%</td><td>22.6%</td></tr><tr><td>π_RLinf</td><td>50.0%</td><td>42.0%</td><td>59.0%</td><td>71.0%</td><td>78.0%</td><td>45.0%</td><td>42.0%</td><td>49.0%</td><td>14.0%</td></tr><tr><td>π0.5</td><td>11.0%</td><td>1.0%</td><td>20.0%</td><td>1.0%</td><td>17.0%</td><td>2.0%</td><td>38.0%</td><td>1.0%</td><td>8.0%</td></tr><tr><td>AtomVLA</td><td>6.3%</td><td>1.0%</td><td>16.0%</td><td>0.0%</td><td>10.0%</td><td>11.0%</td><td>2.0%</td><td>9.0%</td><td>1.0%</td></tr><tr><td>X-VLA</td><td>3.8%</td><td>0.0%</td><td>0.0%</td><td>8.0%</td><td>2.0%</td><td>9.0%</td><td>1.0%</td><td>10.0%</td><td>0.0%</td></tr><tr><td>MolmoAct</td><td>1.5%</td><td>0.0%</td><td>0.0%</td><td>0.0%</td><td>6.0%</td><td>0.0%</td><td>0.0%</td><td>6.0%</td><td>0.0%</td></tr><tr><td>π0</td><td>0.3%</td><td>0.0%</td><td>0.0%</td><td>0.0%</td><td>2.0%</td><td>0.0%</td><td>0.0%</td><td>0.0%</td><td>0.0%</td></tr><tr><td>Cap-X</td><td>—</td><td>14.0%</td><td>12.0%</td><td>18.0%</td><td>22.0%</td><td>17.0%</td><td>26.0%</td><td>—</td><td>—</td></tr><tr><td>RATS</td><td>—</td><td>31.0%</td><td>29.0%</td><td>63.0%</td><td>61.0%</td><td>36.0%</td><td>43.0%</td><td>—</td><td>—</td></tr></tbody></table></div><p>ASPIRE: Long Task and Long Swap use zero-shot transfer from the LIBERO-90 skill library.</p><p id="libero-pro-astra-memory" class="memory-context">GPT-6 Astra: Long Task/Swap and the other six suites use separate memory-file snapshots frozen after their respective exploration phases, with no updates during evaluation. Overall combines two non-overlapping batches: Long 157/200 plus the other suites 584/600, giving 741/800 (92.63%); the 800 episodes do not share a single memory snapshot.</p></section>
+       <section data-benchmark="standard-libero"><h2>LIBERO</h2><table><thead><tr><th>Method / model</th><th>Success rate</th></tr></thead><tbody><tr><td>AtomVLA</td><td>97.0%</td></tr><tr><td>Claude Code / Opus-4.7 / max.reasoning</td><td>96.0%</td></tr><tr><td>π_RLinf</td><td>95.3%</td></tr><tr><td>π0</td><td>94.2%</td></tr><tr><td>NORA</td><td>79.5%</td></tr><tr><td>OpenVLA</td><td>76.5%</td></tr></tbody></table></section>
+       <section data-benchmark="robocasa"><h2>RoboCasa365 · Target50</h2><table><thead><tr><th>Method / model</th><th>Success rate</th></tr></thead><tbody><tr><td>Codex / GPT-6 Astra / low / reasoning</td><td>59.20%</td></tr><tr><td>Xiaomi-Robotics-1</td><td>57.4%</td></tr><tr><td>Codex / GPT-5.5 / xhigh / reasoning</td><td>57.1%</td></tr><tr><td>Claude Code / Opus-4.7 / max.reasoning</td><td>48.6%</td></tr><tr><td>WorldDreamer</td><td>35.3%</td></tr><tr><td>RLDX-1</td><td>30.0%</td></tr><tr><td>π0.5</td><td>16.9%</td></tr><tr><td>π0</td><td>14.8%</td></tr></tbody></table><h3>RoboCasa365 · Target50 · All methods &amp; reported scores</h3><div class="table-wrap" tabindex="0"><table><thead><tr><th>Method / model</th><th>Overall</th><th>Atomic-Seen</th><th>Composite-Seen</th><th>Composite-Unseen</th></tr></thead><tbody><tr><td>GPT-6 Astra</td><td>59.20%</td><td>87.78%</td><td>43.75%</td><td>42.50%</td></tr><tr><td>Xiaomi-Robotics-1</td><td>57.4%</td><td>80.2%</td><td>57.1%</td><td>32.1%</td></tr><tr><td>GPT-5.5</td><td>57.1%</td><td>92.0%</td><td>61.0%</td><td>13.8%</td></tr><tr><td>Opus-4.7 / max.reasoning</td><td>48.6%</td><td>79.4%</td><td>47.5%</td><td>15.0%</td></tr><tr><td>WorldDreamer</td><td>35.3%</td><td>66.3%</td><td>26.7%</td><td>9.0%</td></tr><tr><td>RLDX-1</td><td>30.0%</td><td>60.0%</td><td>21.3%</td><td>5.0%</td></tr><tr><td>π0.5</td><td>16.9%</td><td>39.6%</td><td>7.1%</td><td>1.2%</td></tr><tr><td>π0</td><td>14.8%</td><td>34.6%</td><td>6.1%</td><td>1.1%</td></tr></tbody></table></div></section>
+       <section data-benchmark="robotwin"><h2>RoboTwin</h2><table><thead><tr><th>Method / model</th><th>Success rate</th></tr></thead><tbody><tr><td>Codex / GPT-5.5 / xhigh / reasoning</td><td>62.4%</td></tr><tr><td>Claude Code / Opus-4.7 / max.reasoning</td><td>58.4%</td></tr><tr><td>LingBot-VLA</td><td>50.4%</td></tr><tr><td>π0.5</td><td>47.9%</td></tr><tr><td>GR00T-N1.7</td><td>20.7%</td></tr><tr><td>StarVLA</td><td>10.6%</td></tr></tbody></table></section>
+     </div>
+   </div>

docs/source-en/rst_source/overview.rst

@@ -26,6 +26,26 @@ Together, these principles allow RPent to move beyond traditional robot control
 and establish an agentic infrastructure for the physical world, where intelligence 
 is not only deployed, but continuously built, expanded, and evolved.
 
+Benchmark Results
+-----------------
+
+Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin
+C2R. Rankings apply to the methods and evaluation coverage shown; see
+:doc:`benchmarks` for detailed results, configurations, and sources.
+
+.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard-en-light.png
+   :alt: RPent benchmark results
+   :class: only-light
+   :width: 100%
+   :target: benchmarks.html
+
+.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard-en-dark.png
+   :alt: RPent benchmark results
+   :class: only-dark
+   :width: 100%
+   :target: benchmarks.html
+
+
 Feature Matrix
 --------------
 

docs/source-en/rst_source/usage/libero.rst

@@ -278,11 +278,16 @@ See :doc:`../development/add_primitive` for the full walkthrough.
 Reproducing results
 -------------------
 
-The following results reproduce
-:doc:`Harness VLA <../awesome_works/harnessvla>` on two LIBERO-PRO suites.
-On the `reproduce/libero
-<https://github.com/RLinf/RPent/tree/reproduce/libero>`_ branch, use
-``gpt-5.5`` to reproduce these results:
+See :doc:`../benchmarks` for the unified RPent model comparison on LIBERO-PRO
+Task/Swap and the corresponding model configurations.
+
+The :doc:`GPT-6 Astra suite results <../benchmarks>`
+cover all eight complete suites and 800 verified episodes: 741 successes,
+59 failures, and 92.63% Overall, with Codex / GPT-6 Astra / low / reasoning.
+
+The following historical reproduction records use the `reproduce/libero
+<https://github.com/RLinf/RPent/tree/reproduce/libero>`_ branch with
+``gpt-5.5`` and ``xhigh`` reasoning effort:
 
 - ``libero_10_task``: 70% (70/100)
 - ``libero_10_swap``: 55% (55/100)

docs/source-zh/index.rst

@@ -66,6 +66,7 @@
    概览 <rst_source/overview>
    安装 <rst_source/installation>
    快速开始 <rst_source/quickstart>
+   基准测试结果 <rst_source/benchmarks>
 
 .. toctree::
    :maxdepth: 2

docs/source-zh/rst_source/benchmarks.rst

@@ -0,0 +1,24 @@
+:html_theme.sidebar_secondary.remove:
+
+.. _benchmark-results:
+.. _benchmark-leaderboard:
+.. _leaderboard:
+
+RPent 排行榜
+=================
+
+.. raw:: html
+
+   <link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/docs.css">
+   <script defer src="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/table-sort.js"></script>
+   <script defer src="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard.js"></script>
+   <script defer src="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/embed.js"></script>
+   <div id="rpent-interactive-leaderboard" data-language="zh"
+        data-results-url="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/results.json">
+     <div class="rpent-static-leaderboard">
+       <section data-benchmark="libero-pro"><h2>LIBERO-PRO</h2><table><thead><tr><th>方法 / 模型</th><th>成功率</th></tr></thead><tbody><tr><td>Codex / GPT-6 Astra / low / reasoning</td><td>92.63%</td></tr><tr><td>Claude Code / Opus-4.7 / max.reasoning</td><td>82.4%</td></tr><tr><td>RPent Flash Mode</td><td>72.63%</td></tr><tr><td>Codex / GPT-5.5 / xhigh / reasoning</td><td>72.1%</td></tr><tr><td>ASPIRE</td><td>61.36%</td></tr><tr><td>π_RLinf</td><td>50.0%</td></tr><tr><td>π0.5</td><td>11.0%</td></tr><tr><td>AtomVLA</td><td>6.3%</td></tr><tr><td>X-VLA</td><td>3.8%</td></tr><tr><td>MolmoAct</td><td>1.5%</td></tr><tr><td>π0</td><td>0.3%</td></tr></tbody></table><p>GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).</p><h3>完整方法与分项成绩</h3><div class="table-wrap" tabindex="0"><table><thead><tr><th>方法 / 模型</th><th>总体</th><th>Spatial Task</th><th>Spatial Swap</th><th>Object Task</th><th>Object Swap</th><th>Goal Task</th><th>Goal Swap</th><th>Long Task</th><th>Long Swap</th></tr></thead><tbody><tr><td>GPT-6 Astra</td><td>92.63%</td><td>100%</td><td>98%</td><td>100%</td><td>99%</td><td>88%</td><td>99%</td><td>85%</td><td>72%</td></tr><tr><td>Opus-4.7 / max.reasoning</td><td>82.4%</td><td>—</td><td>—</td><td>—</td><td>—</td><td>—</td><td>—</td><td>—</td><td>—</td></tr><tr><td>RPent Flash Mode / Molmo2-8B</td><td>72.63%</td><td>79.00%</td><td>70.00%</td><td>86.00%</td><td>93.00%</td><td>74.00%</td><td>65.00%</td><td>60.00%</td><td>54.00%</td></tr><tr><td>GPT-5.5</td><td>72.1%</td><td>81.0%</td><td>69.0%</td><td>94.0%</td><td>91.0%</td><td>75.0%</td><td>66.0%</td><td>52.0%</td><td>49.0%</td></tr><tr><td>ASPIRE</td><td>61.36%</td><td>60.0%</td><td>51.0%</td><td>95.0%</td><td>98.0%</td><td>45.0%</td><td>81.0%</td><td>38.3%</td><td>22.6%</td></tr><tr><td>π_RLinf</td><td>50.0%</td><td>42.0%</td><td>59.0%</td><td>71.0%</td><td>78.0%</td><td>45.0%</td><td>42.0%</td><td>49.0%</td><td>14.0%</td></tr><tr><td>π0.5</td><td>11.0%</td><td>1.0%</td><td>20.0%</td><td>1.0%</td><td>17.0%</td><td>2.0%</td><td>38.0%</td><td>1.0%</td><td>8.0%</td></tr><tr><td>AtomVLA</td><td>6.3%</td><td>1.0%</td><td>16.0%</td><td>0.0%</td><td>10.0%</td><td>11.0%</td><td>2.0%</td><td>9.0%</td><td>1.0%</td></tr><tr><td>X-VLA</td><td>3.8%</td><td>0.0%</td><td>0.0%</td><td>8.0%</td><td>2.0%</td><td>9.0%</td><td>1.0%</td><td>10.0%</td><td>0.0%</td></tr><tr><td>MolmoAct</td><td>1.5%</td><td>0.0%</td><td>0.0%</td><td>0.0%</td><td>6.0%</td><td>0.0%</td><td>0.0%</td><td>6.0%</td><td>0.0%</td></tr><tr><td>π0</td><td>0.3%</td><td>0.0%</td><td>0.0%</td><td>0.0%</td><td>2.0%</td><td>0.0%</td><td>0.0%</td><td>0.0%</td><td>0.0%</td></tr><tr><td>Cap-X</td><td>—</td><td>14.0%</td><td>12.0%</td><td>18.0%</td><td>22.0%</td><td>17.0%</td><td>26.0%</td><td>—</td><td>—</td></tr><tr><td>RATS</td><td>—</td><td>31.0%</td><td>29.0%</td><td>63.0%</td><td>61.0%</td><td>36.0%</td><td>43.0%</td><td>—</td><td>—</td></tr></tbody></table></div><p>ASPIRE：Long Task 和 Long Swap 使用 LIBERO-90 技能库进行 zero-shot 迁移。</p><p id="libero-pro-astra-memory" class="memory-context">GPT-6 Astra：Long Task/Swap 与其余六套件使用各自探索后冻结的 memory 文件快照，评测期间不更新。Overall 合并两个不重叠批次：Long 157/200，加上其余套件 584/600，得到 741/800（92.63%）；并非全部回合共享同一份 memory 快照。</p></section>
+       <section data-benchmark="standard-libero"><h2>LIBERO</h2><table><thead><tr><th>方法 / 模型</th><th>成功率</th></tr></thead><tbody><tr><td>AtomVLA</td><td>97.0%</td></tr><tr><td>Claude Code / Opus-4.7 / max.reasoning</td><td>96.0%</td></tr><tr><td>π_RLinf</td><td>95.3%</td></tr><tr><td>π0</td><td>94.2%</td></tr><tr><td>NORA</td><td>79.5%</td></tr><tr><td>OpenVLA</td><td>76.5%</td></tr></tbody></table></section>
+       <section data-benchmark="robocasa"><h2>RoboCasa365 · Target50</h2><table><thead><tr><th>方法 / 模型</th><th>成功率</th></tr></thead><tbody><tr><td>Codex / GPT-6 Astra / low / reasoning</td><td>59.20%</td></tr><tr><td>Xiaomi-Robotics-1</td><td>57.4%</td></tr><tr><td>Codex / GPT-5.5 / xhigh / reasoning</td><td>57.1%</td></tr><tr><td>Claude Code / Opus-4.7 / max.reasoning</td><td>48.6%</td></tr><tr><td>WorldDreamer</td><td>35.3%</td></tr><tr><td>RLDX-1</td><td>30.0%</td></tr><tr><td>π0.5</td><td>16.9%</td></tr><tr><td>π0</td><td>14.8%</td></tr></tbody></table><h3>RoboCasa365 · Target50 · 完整方法与分项成绩</h3><div class="table-wrap" tabindex="0"><table><thead><tr><th>方法 / 模型</th><th>总体</th><th>Atomic-Seen</th><th>Composite-Seen</th><th>Composite-Unseen</th></tr></thead><tbody><tr><td>GPT-6 Astra</td><td>59.20%</td><td>87.78%</td><td>43.75%</td><td>42.50%</td></tr><tr><td>Xiaomi-Robotics-1</td><td>57.4%</td><td>80.2%</td><td>57.1%</td><td>32.1%</td></tr><tr><td>GPT-5.5</td><td>57.1%</td><td>92.0%</td><td>61.0%</td><td>13.8%</td></tr><tr><td>Opus-4.7 / max.reasoning</td><td>48.6%</td><td>79.4%</td><td>47.5%</td><td>15.0%</td></tr><tr><td>WorldDreamer</td><td>35.3%</td><td>66.3%</td><td>26.7%</td><td>9.0%</td></tr><tr><td>RLDX-1</td><td>30.0%</td><td>60.0%</td><td>21.3%</td><td>5.0%</td></tr><tr><td>π0.5</td><td>16.9%</td><td>39.6%</td><td>7.1%</td><td>1.2%</td></tr><tr><td>π0</td><td>14.8%</td><td>34.6%</td><td>6.1%</td><td>1.1%</td></tr></tbody></table></div></section>
+       <section data-benchmark="robotwin"><h2>RoboTwin</h2><table><thead><tr><th>方法 / 模型</th><th>成功率</th></tr></thead><tbody><tr><td>Codex / GPT-5.5 / xhigh / reasoning</td><td>62.4%</td></tr><tr><td>Claude Code / Opus-4.7 / max.reasoning</td><td>58.4%</td></tr><tr><td>LingBot-VLA</td><td>50.4%</td></tr><tr><td>π0.5</td><td>47.9%</td></tr><tr><td>GR00T-N1.7</td><td>20.7%</td></tr><tr><td>StarVLA</td><td>10.6%</td></tr></tbody></table></section>
+     </div>
+   </div>