Scale Overall bars from 70 to 100

index.html

@@ -122,7 +122,7 @@ footer{border-top:1px solid var(--line);color:var(--faint);font-size:12.5px;padd
 </main>
 <footer><div class="wrap"><span data-i18n="foot"></span> · <a href="https://keep-tech.github.io/MoveBench/">keep-tech.github.io/MoveBench</a></div></footer>
 <script>
-const M=[{"rank": 1, "name": "GPT-5.2", "cov": "全量", "total": 0.924, "oq": 0.895, "mcq": 0.971, "DA": 0.94, "PG": 0.872, "QA": 0.957, "ss": 1.0, "ms": 0.954, "safe": 0.891, "pro": 0.922, "exp": 0.909, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "GPT", "hf": null}, {"rank": 2, "name": "GLM-5.3", "cov": "全量", "total": 0.922, "oq": 0.889, "mcq": 0.974, "DA": 0.935, "PG": 0.873, "QA": 0.955, "ss": 1.0, "ms": 0.959, "safe": 0.905, "pro": 0.913, "exp": 0.907, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "GLM", "hf": "zai-org/GLM-5.3"}, {"rank": 3, "name": "Claude Opus 5.5", "cov": "全量", "total": 0.921, "oq": 0.895, "mcq": 0.965, "DA": 0.926, "PG": 0.88, "QA": 0.957, "ss": 1.0, "ms": 0.944, "safe": 0.916, "pro": 0.919, "exp": 0.915, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Claude", "hf": null}, {"rank": 4, "name": "Claude 4.6 Opus", "cov": "全量", "total": 0.912, "oq": 0.87, "mcq": 0.981, "DA": 0.925, "PG": 0.859, "QA": 0.95, "ss": 1.0, "ms": 0.97, "safe": 0.856, "pro": 0.909, "exp": 0.895, "n": 809, "nOq": 499, "nMcq": 310, "zero": 1, "family": "Claude", "hf": null}, {"rank": 5, "name": "Qwen3.8 Max", "cov": "全量", "total": 0.912, "oq": 0.868, "mcq": 0.984, "DA": 0.918, "PG": 0.881, "QA": 0.936, "ss": 1.0, "ms": 0.975, "safe": 0.915, "pro": 0.88, "exp": 0.9, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Qwen", "hf": null}, {"rank": 6, "name": "Doubao Seed 2.1 Pro", "cov": "全量", "total": 0.91, "oq": 0.861, "mcq": 0.987, "DA": 0.907, "PG": 0.882, "QA": 0.94, "ss": 1.0, "ms": 0.98, "safe": 0.892, "pro": 0.879, "exp": 0.898, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Doubao", "hf": null}, {"rank": 7, "name": "Kimi K3", "cov": "全量", "total": 0.909, "oq": 0.865, "mcq": 0.981, "DA": 0.922, "PG": 0.863, "QA": 0.941, "ss": 1.0, "ms": 0.97, "safe": 0.895, "pro": 0.89, "exp": 0.891, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Kimi", "hf": "moonshotai/Kimi-K3"}, {"rank": 8, "name": "Qwen3.5 397B-A17B", "cov": "全量", "total": 0.909, "oq": 0.872, "mcq": 0.968, "DA": 0.929, "PG": 0.83, "QA": 0.964, "ss": 1.0, "ms": 0.949, "safe": 0.894, "pro": 0.91, "exp": 0.901, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Qwen", "hf": "Qwen/Qwen3.5-397B-A17B"}, {"rank": 9, "name": "MiniMax M3", "cov": "全量", "total": 0.907, "oq": 0.874, "mcq": 0.961, "DA": 0.927, "PG": 0.838, "QA": 0.954, "ss": 1.0, "ms": 0.939, "safe": 0.926, "pro": 0.909, "exp": 0.879, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "MiniMax", "hf": "MiniMaxAI/MiniMax-M3"}, {"rank": 10, "name": "Qwen3.5 27B", "cov": "全量", "total": 0.899, "oq": 0.856, "mcq": 0.968, "DA": 0.925, "PG": 0.808, "QA": 0.96, "ss": 1.0, "ms": 0.949, "safe": 0.869, "pro": 0.897, "exp": 0.894, "n": 809, "nOq": 499, "nMcq": 310, "zero": 1, "family": "Qwen", "hf": "Qwen/Qwen3.5-27B"}, {"rank": 11, "name": "MiMo v2.6 Pro", "cov": "全量", "total": 0.898, "oq": 0.85, "mcq": 0.974, "DA": 0.917, "PG": 0.855, "QA": 0.918, "ss": 1.0, "ms": 0.959, "safe": 0.869, "pro": 0.872, "exp": 0.876, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "MiMo", "hf": null}, {"rank": 12, "name": "Qwen3.5 35B-A3B", "cov": "全量", "total": 0.891, "oq": 0.849, "mcq": 0.958, "DA": 0.921, "PG": 0.799, "QA": 0.948, "ss": 1.0, "ms": 0.934, "safe": 0.858, "pro": 0.889, "exp": 0.885, "n": 809, "nOq": 499, "nMcq": 310, "zero": 3, "family": "Qwen", "hf": "Qwen/Qwen3.5-35B-A3B"}, {"rank": 13, "name": "Gemini 4 Argon", "cov": "全量", "total": 0.89, "oq": 0.83, "mcq": 0.987, "DA": 0.918, "PG": 0.84, "QA": 0.908, "ss": 1.0, "ms": 0.98, "safe": 0.824, "pro": 0.874, "exp": 0.867, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Gemini", "hf": null}, {"rank": 14, "name": "HY3", "cov": "全量", "total": 0.887, "oq": 0.845, "mcq": 0.955, "DA": 0.909, "PG": 0.826, "QA": 0.924, "ss": 1.0, "ms": 0.929, "safe": 0.868, "pro": 0.886, "exp": 0.867, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "HY", "hf": "tencent/Hy3"}, {"rank": 15, "name": "GPT-6", "cov": "全量", "total": 0.88, "oq": 0.846, "mcq": 0.935, "DA": 0.885, "PG": 0.858, "QA": 0.898, "ss": 1.0, "ms": 0.898, "safe": 0.889, "pro": 0.869, "exp": 0.873, "n": 808, "nOq": 499, "nMcq": 309, "zero": 0, "family": "GPT", "hf": null}, {"rank": 16, "name": "Qwen3 Max", "cov": "全量", "total": 0.873, "oq": 0.806, "mcq": 0.981, "DA": 0.905, "PG": 0.804, "QA": 0.906, "ss": 1.0, "ms": 0.97, "safe": 0.772, "pro": 0.851, "exp": 0.861, "n": 809, "nOq": 499, "nMcq": 310, "zero": 2, "family": "Qwen", "hf": null}, {"rank": 17, "name": "DeepSeek R1", "cov": "全量", "total": 0.861, "oq": 0.797, "mcq": 0.965, "DA": 0.892, "PG": 0.78, "QA": 0.907, "ss": 1.0, "ms": 0.944, "safe": 0.788, "pro": 0.84, "exp": 0.852, "n": 809, "nOq": 499, "nMcq": 310, "zero": 2, "family": "DeepSeek", "hf": "deepseek-ai/DeepSeek-R1"}, {"rank": 18, "name": "Doubao Seed 2.0 Pro", "cov": "全量", "total": 0.856, "oq": 0.786, "mcq": 0.968, "DA": 0.877, "PG": 0.783, "QA": 0.903, "ss": 1.0, "ms": 0.949, "safe": 0.784, "pro": 0.836, "exp": 0.849, "n": 809, "nOq": 499, "nMcq": 310, "zero": 2, "family": "Doubao", "hf": null}, {"rank": 19, "name": "Kimi K2", "cov": "全量", "total": 0.847, "oq": 0.765, "mcq": 0.981, "DA": 0.895, "PG": 0.737, "QA": 0.903, "ss": 1.0, "ms": 0.97, "safe": 0.753, "pro": 0.83, "exp": 0.822, "n": 809, "nOq": 499, "nMcq": 310, "zero": 6, "family": "Kimi", "hf": null}, {"rank": 20, "name": "GLM-4.7", "cov": "全量", "total": 0.843, "oq": 0.79, "mcq": 0.929, "DA": 0.894, "PG": 0.716, "QA": 0.911, "ss": 1.0, "ms": 0.888, "safe": 0.752, "pro": 0.854, "exp": 0.854, "n": 809, "nOq": 499, "nMcq": 310, "zero": 4, "family": "GLM", "hf": "zai-org/GLM-4.7"}, {"rank": 21, "name": "Grok 4.1 Fast", "cov": "全量", "total": 0.842, "oq": 0.77, "mcq": 0.958, "DA": 0.869, "PG": 0.766, "QA": 0.887, "ss": 1.0, "ms": 0.934, "safe": 0.714, "pro": 0.809, "exp": 0.834, "n": 809, "nOq": 499, "nMcq": 310, "zero": 3, "family": "Grok", "hf": null}, {"rank": 22, "name": "DeepSeek V3.2/DeepSeek V3", "cov": "全量(OQ=V3.2/MCQ=V3)", "total": 0.842, "oq": 0.769, "mcq": 0.958, "DA": 0.847, "PG": 0.787, "QA": 0.89, "ss": 1.0, "ms": 0.934, "safe": 0.737, "pro": 0.808, "exp": 0.832, "n": 809, "nOq": 499, "nMcq": 310, "zero": 8, "family": "DeepSeek", "hf": null}, {"rank": 23, "name": "GPT-OSS 120B", "cov": "全量", "total": 0.81, "oq": 0.723, "mcq": 0.952, "DA": 0.827, "PG": 0.723, "QA": 0.878, "ss": 1.0, "ms": 0.924, "safe": 0.672, "pro": 0.769, "exp": 0.799, "n": 809, "nOq": 499, "nMcq": 310, "zero": 10, "family": "GPT", "hf": "openai/gpt-oss-120b"}, {"rank": 24, "name": "Qwen3 235B-A22B", "cov": "全量", "total": 0.78, "oq": 0.672, "mcq": 0.955, "DA": 0.81, "PG": 0.686, "QA": 0.84, "ss": 1.0, "ms": 0.929, "safe": 0.59, "pro": 0.721, "exp": 0.769, "n": 809, "nOq": 499, "nMcq": 310, "zero": 10, "family": "Qwen", "hf": "Qwen/Qwen3-235B-A22B"}, {"rank": 25, "name": "Qwen3 32B", "cov": "全量", "total": 0.734, "oq": 0.595, "mcq": 0.958, "DA": 0.77, "PG": 0.636, "QA": 0.79, "ss": 1.0, "ms": 0.934, "safe": 0.516, "pro": 0.636, "exp": 0.711, "n": 809, "nOq": 499, "nMcq": 310, "zero": 21, "family": "Qwen", "hf": "Qwen/Qwen3-32B"}],T={"en": {"eyebrow": "Leaderboard · Movement intelligence", "h1a": "MoveBench", "h1b": "Sports Science LLM Leaderboard", "lead": "A unified framework of expertise, safety and experience, measuring what large language models can and cannot do across three real coaching scenarios: workout data analysis, personalized training plans and expert knowledge QA.", "st1": "models evaluated", "st2": "questions · 499 open-ended + 310 multiple-choice", "st3": "scenarios · DA / PG / QA", "st4": "evaluation layers · safety / expertise / experience", "t_overall": "Overall", "t_scene": "Scenarios", "t_cap": "Capability", "t_type": "Question type", "t_prog": "Progress", "t_about": "About", "h_overall": "Overall leaderboard", "s_overall": "Full 809-question evaluation, scores out of 100. Click a column header to sort.", "search": "Search models", "allfam": "All families", "g_fmt": "Format", "g_sc": "Scenario", "g_layer": "Layer", "c_rank": "#", "c_name": "Model", "c_total": "Overall", "c_oq": "OQ", "c_mcq": "MCQ", "c_ss": "SS", "c_safe": "Safety", "c_pro": "Expertise", "c_exp": "Experience", "c_zero": "Zeroed", "n_overall": "OQ = open-ended, MCQ = multiple-choice, SS = single-select; Zeroed = open-ended answers scored 0 for a safety red-line violation. † GPT-6 answered 808 questions (one multiple-choice question missing). ‡ DeepSeek V3.2 / V3 combines V3.2 on open-ended and V3 on multiple-choice questions. Overall scores are taken from the source table, not recomputed from sub-scores. Models marked <span class=\"hfb\">HF</span> link to their Hugging Face model page.", "empty": "No matching models", "h_scene": "Scenario leaderboards", "s_scene": "Each scenario ranked separately. Personalized training plans (PG) are the lowest-scoring scenario for nearly every model.", "DA_t": "Workout data analysis", "DA_d": "Data understanding, metric interpretation and action advice", "PG_t": "Personalized training plans", "PG_d": "Goal fit, course selection and plan structure", "QA_t": "Expert knowledge QA", "QA_d": "Knowledge accuracy, applicability and clarity", "top": "Top", "h_cap": "Capability leaderboard", "s_cap": "Safety, expertise and experience layers side by side.", "so_total": "Sort by overall", "so_safe": "Sort by safety", "so_pro": "Sort by expertise", "so_exp": "Sort by experience", "so_ms": "Sort by multi-select", "so_oq": "Sort by open-ended", "l_safe": "Safety", "l_pro": "Expertise", "l_exp": "Experience", "l_ss": "Single-select", "l_ms": "Multi-select", "l_oq": "Open-ended", "h_type": "Question-type leaderboard", "s_type": "Single-select, multi-select and open-ended side by side. Every model scores 100 on single-select, so open-ended questions do most of the separating.", "h_prog": "Progress leaderboard", "s_prog": "Change in overall score between two versions in the same model family. This is a score difference between versions, not a rate over time, and is not attributable to the version upgrade alone.", "n_prog": "For Qwen, only Max → Max is linked. 32B, 27B, 35B-A3B, 235B-A22B and 397B-A17B differ in size or architecture and are not treated as an upgrade chain.", "rankN": "Gain #", "pts": "pts", "safeL": "Safety", "h_about": "About the evaluation", "s_about": "Framework, scoring and data notes.", "a_fw": "Framework", "p_safe": "Safety", "p_pro": "Expertise", "p_exp": "Experience", "a_fw_l": "<li><b>DA · Workout data analysis</b>: data understanding, metric interpretation and action advice</li><li><b>PG · Personalized training plans</b>: goal fit, course selection and plan structure</li><li><b>QA · Expert knowledge QA</b>: knowledge accuracy, applicability and clarity</li>", "a_sc": "Scoring", "a_sc_l": "<li>Open-ended answers are graded against expert-written, per-question rubrics by a three-judge weighted LLM panel.</li><li>Single- and multi-select questions are scored by exact match against the gold answer.</li><li>An open-ended answer that violates a safety red line scores 0. The \"Zeroed\" column counts these for each model.</li>", "a_dn": "Data notes", "a_dn_l": "<li>Scores are the source scores × 100, rounded to one decimal.</li><li>This page reports the full 809-question evaluation. Scores on the 80-question public subset are on the <a href=\"https://huggingface.co/datasets/KeepaceAI/MoveBench\">dataset page</a>; the two are not interchangeable.</li><li>Model names follow the source table.</li>", "a_sub": "Get your model listed", "a_sub_l": "<li>Download the <a href=\"https://huggingface.co/datasets/KeepaceAI/MoveBench\">dataset</a> and generate answers with your model.</li><li>Submit the answer file following the <a href=\"https://github.com/Keep-Tech/MoveBench/blob/main/docs/SUBMISSION.md\">submission guide</a>. Maintainers score and review submissions before updating the leaderboard.</li><li>Cite the <a href=\"https://arxiv.org/abs/2610.26185\">MoveBench paper</a> (arXiv 2610.26185).</li>", "foot": "MoveBench Team × Capital University of Physical Education and Sports · Dataset and documentation licensed under CC BY 4.0"}, "zh": {"eyebrow": "Leaderboard · Movement intelligence", "h1a": "MoveBench", "h1b": "运动科学大模型榜单", "lead": "以专业、安全、体验为统一评价框架，在运动数据解读、个性化课程方案、专业知识问答三个真实服务场景中衡量大模型的能力与边界。", "st1": "参评模型", "st2": "评测题 · 499 道开放题 + 310 道选择题", "st3": "业务主题 · DA / PG / QA", "st4": "评价层面 · 安全 / 专业 / 体验", "t_overall": "总榜", "t_scene": "主题榜", "t_cap": "能力榜", "t_type": "题型榜", "t_prog": "进步榜", "t_about": "说明", "h_overall": "总榜", "s_overall": "809 题全量评测，分数为百分制。点击表头排序。", "search": "搜索模型", "allfam": "全部系列", "g_fmt": "题型", "g_sc": "主题", "g_layer": "评价层面", "c_rank": "#", "c_name": "模型", "c_total": "总分", "c_oq": "开放题", "c_mcq": "选择题", "c_ss": "单选", "c_safe": "安全", "c_pro": "专业", "c_exp": "体验", "c_zero": "红线置零", "n_overall": "OQ = open-ended, MCQ = multiple-choice, SS = single-select; Zeroed = open-ended answers scored 0 for a safety red-line violation. † GPT-6 为 808 题（少 1 道选择题）。‡ DeepSeek V3.2 / V3：开放题用 V3.2、选择题用 V3 的混合结果。总分沿用原表，不由分项均值重算。带 <span class=\"hfb\">HF</span> 标记的模型可跳转到其 Hugging Face 模型页。", "empty": "没有匹配的模型", "h_scene": "主题榜", "s_scene": "三个业务场景各自排名。几乎所有模型都在 PG 个性化课程方案上得分最低。", "DA_t": "运动数据解读", "DA_d": "数据理解、指标解释与行动建议", "PG_t": "个性化课程方案", "PG_d": "目标适配、课程选择与方案编排", "QA_t": "专业知识问答", "QA_d": "知识准确、适用条件与表达质量", "top": "最高", "h_cap": "能力榜", "s_cap": "安全层、专业层、体验层并排对照。", "so_total": "按总分排序", "so_safe": "按安全层排序", "so_pro": "按专业层排序", "so_exp": "按体验层排序", "so_ms": "按多选排序", "so_oq": "按开放问答排序", "l_safe": "安全层", "l_pro": "专业层", "l_exp": "体验层", "l_ss": "单选", "l_ms": "多选", "l_oq": "开放问答", "h_type": "题型榜", "s_type": "单选、多选、开放问答并排对照。单选题全部满分，区分度主要来自开放问答。", "h_prog": "进步榜", "s_prog": "同系列两个版本之间的总分变化。这里的“进步”指版本间分数差，不代表时间速度，也不能单独归因于版本升级。", "n_prog": "Qwen 系列只连接 Max → Max；32B、27B、35B-A3B、235B-A22B、397B-A17B 属于不同规模或架构，不作为连续升级链。", "rankN": "增幅第 ", "pts": "分", "safeL": "安全", "h_about": "评测说明", "s_about": "评价框架、评分方式与数据口径。", "a_fw": "评价框架", "p_safe": "安全", "p_pro": "专业", "p_exp": "体验", "a_fw_l": "<li><b>DA 运动数据解读</b>：数据理解、指标解释与行动建议</li><li><b>PG 个性化课程方案</b>：目标适配、课程选择与方案编排</li><li><b>QA 专业知识问答</b>：知识准确、适用条件与表达质量</li>", "a_sc": "评分方式", "a_sc_l": "<li>开放题由三个 LLM judge 组成的加权评审团，依据专家编写的逐题 rubric 评分。</li><li>选择题（单选、多选）按标准答案精确匹配评分。</li><li>触犯安全红线的开放题记 0 分；总榜“红线置零”列为每个模型被置零的题数。</li>", "a_dn": "数据口径", "a_dn_l": "<li>分数为原表分数 × 100，保留 1 位小数。</li><li>本页为 809 题全量评测结果。公开的 80 题子集成绩见<a href=\"https://huggingface.co/datasets/KeepaceAI/MoveBench\">数据集页</a>，两者不可混用。</li><li>模型名称沿用原表写法。</li>", "a_sub": "参与评测", "a_sub_l": "<li>下载<a href=\"https://huggingface.co/datasets/KeepaceAI/MoveBench\">数据集</a>，用你的模型生成答案。</li><li>按<a href=\"https://github.com/Keep-Tech/MoveBench/blob/main/docs/SUBMISSION.md\">提交指南</a>提交答案文件，维护方评分审核后更新榜单。</li><li>引用：<a href=\"https://arxiv.org/abs/2610.26185\">MoveBench 论文</a>（arXiv 2610.26185）。</li>", "foot": "MoveBench Team × 首都体育学院 · 数据集与文档采用 CC BY 4.0 许可"}};
+const M=[{"rank": 1, "name": "GPT-5.2", "cov": "全量", "total": 0.924, "oq": 0.895, "mcq": 0.971, "DA": 0.94, "PG": 0.872, "QA": 0.957, "ss": 1.0, "ms": 0.954, "safe": 0.891, "pro": 0.922, "exp": 0.909, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "GPT", "hf": null}, {"rank": 2, "name": "GLM-5.3", "cov": "全量", "total": 0.922, "oq": 0.889, "mcq": 0.974, "DA": 0.935, "PG": 0.873, "QA": 0.955, "ss": 1.0, "ms": 0.959, "safe": 0.905, "pro": 0.913, "exp": 0.907, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "GLM", "hf": "zai-org/GLM-5.3"}, {"rank": 3, "name": "Claude Opus 5.5", "cov": "全量", "total": 0.921, "oq": 0.895, "mcq": 0.965, "DA": 0.926, "PG": 0.88, "QA": 0.957, "ss": 1.0, "ms": 0.944, "safe": 0.916, "pro": 0.919, "exp": 0.915, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Claude", "hf": null}, {"rank": 4, "name": "Claude 4.6 Opus", "cov": "全量", "total": 0.912, "oq": 0.87, "mcq": 0.981, "DA": 0.925, "PG": 0.859, "QA": 0.95, "ss": 1.0, "ms": 0.97, "safe": 0.856, "pro": 0.909, "exp": 0.895, "n": 809, "nOq": 499, "nMcq": 310, "zero": 1, "family": "Claude", "hf": null}, {"rank": 5, "name": "Qwen3.8 Max", "cov": "全量", "total": 0.912, "oq": 0.868, "mcq": 0.984, "DA": 0.918, "PG": 0.881, "QA": 0.936, "ss": 1.0, "ms": 0.975, "safe": 0.915, "pro": 0.88, "exp": 0.9, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Qwen", "hf": null}, {"rank": 6, "name": "Doubao Seed 2.1 Pro", "cov": "全量", "total": 0.91, "oq": 0.861, "mcq": 0.987, "DA": 0.907, "PG": 0.882, "QA": 0.94, "ss": 1.0, "ms": 0.98, "safe": 0.892, "pro": 0.879, "exp": 0.898, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Doubao", "hf": null}, {"rank": 7, "name": "Kimi K3", "cov": "全量", "total": 0.909, "oq": 0.865, "mcq": 0.981, "DA": 0.922, "PG": 0.863, "QA": 0.941, "ss": 1.0, "ms": 0.97, "safe": 0.895, "pro": 0.89, "exp": 0.891, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Kimi", "hf": "moonshotai/Kimi-K3"}, {"rank": 8, "name": "Qwen3.5 397B-A17B", "cov": "全量", "total": 0.909, "oq": 0.872, "mcq": 0.968, "DA": 0.929, "PG": 0.83, "QA": 0.964, "ss": 1.0, "ms": 0.949, "safe": 0.894, "pro": 0.91, "exp": 0.901, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Qwen", "hf": "Qwen/Qwen3.5-397B-A17B"}, {"rank": 9, "name": "MiniMax M3", "cov": "全量", "total": 0.907, "oq": 0.874, "mcq": 0.961, "DA": 0.927, "PG": 0.838, "QA": 0.954, "ss": 1.0, "ms": 0.939, "safe": 0.926, "pro": 0.909, "exp": 0.879, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "MiniMax", "hf": "MiniMaxAI/MiniMax-M3"}, {"rank": 10, "name": "Qwen3.5 27B", "cov": "全量", "total": 0.899, "oq": 0.856, "mcq": 0.968, "DA": 0.925, "PG": 0.808, "QA": 0.96, "ss": 1.0, "ms": 0.949, "safe": 0.869, "pro": 0.897, "exp": 0.894, "n": 809, "nOq": 499, "nMcq": 310, "zero": 1, "family": "Qwen", "hf": "Qwen/Qwen3.5-27B"}, {"rank": 11, "name": "MiMo v2.6 Pro", "cov": "全量", "total": 0.898, "oq": 0.85, "mcq": 0.974, "DA": 0.917, "PG": 0.855, "QA": 0.918, "ss": 1.0, "ms": 0.959, "safe": 0.869, "pro": 0.872, "exp": 0.876, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "MiMo", "hf": null}, {"rank": 12, "name": "Qwen3.5 35B-A3B", "cov": "全量", "total": 0.891, "oq": 0.849, "mcq": 0.958, "DA": 0.921, "PG": 0.799, "QA": 0.948, "ss": 1.0, "ms": 0.934, "safe": 0.858, "pro": 0.889, "exp": 0.885, "n": 809, "nOq": 499, "nMcq": 310, "zero": 3, "family": "Qwen", "hf": "Qwen/Qwen3.5-35B-A3B"}, {"rank": 13, "name": "Gemini 4 Argon", "cov": "全量", "total": 0.89, "oq": 0.83, "mcq": 0.987, "DA": 0.918, "PG": 0.84, "QA": 0.908, "ss": 1.0, "ms": 0.98, "safe": 0.824, "pro": 0.874, "exp": 0.867, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "Gemini", "hf": null}, {"rank": 14, "name": "HY3", "cov": "全量", "total": 0.887, "oq": 0.845, "mcq": 0.955, "DA": 0.909, "PG": 0.826, "QA": 0.924, "ss": 1.0, "ms": 0.929, "safe": 0.868, "pro": 0.886, "exp": 0.867, "n": 809, "nOq": 499, "nMcq": 310, "zero": 0, "family": "HY", "hf": "tencent/Hy3"}, {"rank": 15, "name": "GPT-6", "cov": "全量", "total": 0.88, "oq": 0.846, "mcq": 0.935, "DA": 0.885, "PG": 0.858, "QA": 0.898, "ss": 1.0, "ms": 0.898, "safe": 0.889, "pro": 0.869, "exp": 0.873, "n": 808, "nOq": 499, "nMcq": 309, "zero": 0, "family": "GPT", "hf": null}, {"rank": 16, "name": "Qwen3 Max", "cov": "全量", "total": 0.873, "oq": 0.806, "mcq": 0.981, "DA": 0.905, "PG": 0.804, "QA": 0.906, "ss": 1.0, "ms": 0.97, "safe": 0.772, "pro": 0.851, "exp": 0.861, "n": 809, "nOq": 499, "nMcq": 310, "zero": 2, "family": "Qwen", "hf": null}, {"rank": 17, "name": "DeepSeek R1", "cov": "全量", "total": 0.861, "oq": 0.797, "mcq": 0.965, "DA": 0.892, "PG": 0.78, "QA": 0.907, "ss": 1.0, "ms": 0.944, "safe": 0.788, "pro": 0.84, "exp": 0.852, "n": 809, "nOq": 499, "nMcq": 310, "zero": 2, "family": "DeepSeek", "hf": "deepseek-ai/DeepSeek-R1"}, {"rank": 18, "name": "Doubao Seed 2.0 Pro", "cov": "全量", "total": 0.856, "oq": 0.786, "mcq": 0.968, "DA": 0.877, "PG": 0.783, "QA": 0.903, "ss": 1.0, "ms": 0.949, "safe": 0.784, "pro": 0.836, "exp": 0.849, "n": 809, "nOq": 499, "nMcq": 310, "zero": 2, "family": "Doubao", "hf": null}, {"rank": 19, "name": "Kimi K2", "cov": "全量", "total": 0.847, "oq": 0.765, "mcq": 0.981, "DA": 0.895, "PG": 0.737, "QA": 0.903, "ss": 1.0, "ms": 0.97, "safe": 0.753, "pro": 0.83, "exp": 0.822, "n": 809, "nOq": 499, "nMcq": 310, "zero": 6, "family": "Kimi", "hf": null}, {"rank": 20, "name": "GLM-4.7", "cov": "全量", "total": 0.843, "oq": 0.79, "mcq": 0.929, "DA": 0.894, "PG": 0.716, "QA": 0.911, "ss": 1.0, "ms": 0.888, "safe": 0.752, "pro": 0.854, "exp": 0.854, "n": 809, "nOq": 499, "nMcq": 310, "zero": 4, "family": "GLM", "hf": "zai-org/GLM-4.7"}, {"rank": 21, "name": "Grok 4.1 Fast", "cov": "全量", "total": 0.842, "oq": 0.77, "mcq": 0.958, "DA": 0.869, "PG": 0.766, "QA": 0.887, "ss": 1.0, "ms": 0.934, "safe": 0.714, "pro": 0.809, "exp": 0.834, "n": 809, "nOq": 499, "nMcq": 310, "zero": 3, "family": "Grok", "hf": null}, {"rank": 22, "name": "DeepSeek V3.2/DeepSeek V3", "cov": "全量(OQ=V3.2/MCQ=V3)", "total": 0.842, "oq": 0.769, "mcq": 0.958, "DA": 0.847, "PG": 0.787, "QA": 0.89, "ss": 1.0, "ms": 0.934, "safe": 0.737, "pro": 0.808, "exp": 0.832, "n": 809, "nOq": 499, "nMcq": 310, "zero": 8, "family": "DeepSeek", "hf": null}, {"rank": 23, "name": "GPT-OSS 120B", "cov": "全量", "total": 0.81, "oq": 0.723, "mcq": 0.952, "DA": 0.827, "PG": 0.723, "QA": 0.878, "ss": 1.0, "ms": 0.924, "safe": 0.672, "pro": 0.769, "exp": 0.799, "n": 809, "nOq": 499, "nMcq": 310, "zero": 10, "family": "GPT", "hf": "openai/gpt-oss-120b"}, {"rank": 24, "name": "Qwen3 235B-A22B", "cov": "全量", "total": 0.78, "oq": 0.672, "mcq": 0.955, "DA": 0.81, "PG": 0.686, "QA": 0.84, "ss": 1.0, "ms": 0.929, "safe": 0.59, "pro": 0.721, "exp": 0.769, "n": 809, "nOq": 499, "nMcq": 310, "zero": 10, "family": "Qwen", "hf": "Qwen/Qwen3-235B-A22B"}, {"rank": 25, "name": "Qwen3 32B", "cov": "全量", "total": 0.734, "oq": 0.595, "mcq": 0.958, "DA": 0.77, "PG": 0.636, "QA": 0.79, "ss": 1.0, "ms": 0.934, "safe": 0.516, "pro": 0.636, "exp": 0.711, "n": 809, "nOq": 499, "nMcq": 310, "zero": 21, "family": "Qwen", "hf": "Qwen/Qwen3-32B"}],T={"en": {"eyebrow": "Leaderboard · Movement intelligence", "h1a": "MoveBench", "h1b": "Sports Science LLM Leaderboard", "lead": "A unified framework of expertise, safety and experience, measuring what large language models can and cannot do across three real coaching scenarios: workout data analysis, personalized training plans and expert knowledge QA.", "st1": "models evaluated", "st2": "questions · 499 open-ended + 310 multiple-choice", "st3": "scenarios · DA / PG / QA", "st4": "evaluation layers · safety / expertise / experience", "t_overall": "Overall", "t_scene": "Scenarios", "t_cap": "Capability", "t_type": "Question type", "t_prog": "Progress", "t_about": "About", "h_overall": "Overall leaderboard", "s_overall": "Full 809-question evaluation, scores out of 100. Click a column header to sort.", "search": "Search models", "allfam": "All families", "g_fmt": "Format", "g_sc": "Scenario", "g_layer": "Layer", "c_rank": "#", "c_name": "Model", "c_total": "Overall", "c_oq": "OQ", "c_mcq": "MCQ", "c_ss": "SS", "c_safe": "Safety", "c_pro": "Expertise", "c_exp": "Experience", "c_zero": "Zeroed", "n_overall": "OQ = open-ended, MCQ = multiple-choice, SS = single-select; Zeroed = open-ended answers scored 0 for a safety red-line violation. † GPT-6 answered 808 questions (one multiple-choice question missing). ‡ DeepSeek V3.2 / V3 combines V3.2 on open-ended and V3 on multiple-choice questions. Overall scores are taken from the source table, not recomputed from sub-scores. Bars in the Overall column run from 70 to 100 to make the gaps visible. Models marked <span class=\"hfb\">HF</span> link to their Hugging Face model page.", "empty": "No matching models", "h_scene": "Scenario leaderboards", "s_scene": "Each scenario ranked separately. Personalized training plans (PG) are the lowest-scoring scenario for nearly every model.", "DA_t": "Workout data analysis", "DA_d": "Data understanding, metric interpretation and action advice", "PG_t": "Personalized training plans", "PG_d": "Goal fit, course selection and plan structure", "QA_t": "Expert knowledge QA", "QA_d": "Knowledge accuracy, applicability and clarity", "top": "Top", "h_cap": "Capability leaderboard", "s_cap": "Safety, expertise and experience layers side by side.", "so_total": "Sort by overall", "so_safe": "Sort by safety", "so_pro": "Sort by expertise", "so_exp": "Sort by experience", "so_ms": "Sort by multi-select", "so_oq": "Sort by open-ended", "l_safe": "Safety", "l_pro": "Expertise", "l_exp": "Experience", "l_ss": "Single-select", "l_ms": "Multi-select", "l_oq": "Open-ended", "h_type": "Question-type leaderboard", "s_type": "Single-select, multi-select and open-ended side by side. Every model scores 100 on single-select, so open-ended questions do most of the separating.", "h_prog": "Progress leaderboard", "s_prog": "Change in overall score between two versions in the same model family. This is a score difference between versions, not a rate over time, and is not attributable to the version upgrade alone.", "n_prog": "For Qwen, only Max → Max is linked. 32B, 27B, 35B-A3B, 235B-A22B and 397B-A17B differ in size or architecture and are not treated as an upgrade chain.", "rankN": "Gain #", "pts": "pts", "safeL": "Safety", "h_about": "About the evaluation", "s_about": "Framework, scoring and data notes.", "a_fw": "Framework", "p_safe": "Safety", "p_pro": "Expertise", "p_exp": "Experience", "a_fw_l": "<li><b>DA · Workout data analysis</b>: data understanding, metric interpretation and action advice</li><li><b>PG · Personalized training plans</b>: goal fit, course selection and plan structure</li><li><b>QA · Expert knowledge QA</b>: knowledge accuracy, applicability and clarity</li>", "a_sc": "Scoring", "a_sc_l": "<li>Open-ended answers are graded against expert-written, per-question rubrics by a three-judge weighted LLM panel.</li><li>Single- and multi-select questions are scored by exact match against the gold answer.</li><li>An open-ended answer that violates a safety red line scores 0. The \"Zeroed\" column counts these for each model.</li>", "a_dn": "Data notes", "a_dn_l": "<li>Scores are the source scores × 100, rounded to one decimal.</li><li>This page reports the full 809-question evaluation. Scores on the 80-question public subset are on the <a href=\"https://huggingface.co/datasets/KeepaceAI/MoveBench\">dataset page</a>; the two are not interchangeable.</li><li>Model names follow the source table.</li>", "a_sub": "Get your model listed", "a_sub_l": "<li>Download the <a href=\"https://huggingface.co/datasets/KeepaceAI/MoveBench\">dataset</a> and generate answers with your model.</li><li>Submit the answer file following the <a href=\"https://github.com/Keep-Tech/MoveBench/blob/main/docs/SUBMISSION.md\">submission guide</a>. Maintainers score and review submissions before updating the leaderboard.</li><li>Cite the <a href=\"https://arxiv.org/abs/2610.26185\">MoveBench paper</a> (arXiv 2610.26185).</li>", "foot": "MoveBench Team × Capital University of Physical Education and Sports · Dataset and documentation licensed under CC BY 4.0"}, "zh": {"eyebrow": "Leaderboard · Movement intelligence", "h1a": "MoveBench", "h1b": "运动科学大模型榜单", "lead": "以专业、安全、体验为统一评价框架，在运动数据解读、个性化课程方案、专业知识问答三个真实服务场景中衡量大模型的能力与边界。", "st1": "参评模型", "st2": "评测题 · 499 道开放题 + 310 道选择题", "st3": "业务主题 · DA / PG / QA", "st4": "评价层面 · 安全 / 专业 / 体验", "t_overall": "总榜", "t_scene": "主题榜", "t_cap": "能力榜", "t_type": "题型榜", "t_prog": "进步榜", "t_about": "说明", "h_overall": "总榜", "s_overall": "809 题全量评测，分数为百分制。点击表头排序。", "search": "搜索模型", "allfam": "全部系列", "g_fmt": "题型", "g_sc": "主题", "g_layer": "评价层面", "c_rank": "#", "c_name": "模型", "c_total": "总分", "c_oq": "开放题", "c_mcq": "选择题", "c_ss": "单选", "c_safe": "安全", "c_pro": "专业", "c_exp": "体验", "c_zero": "红线置零", "n_overall": "OQ = open-ended, MCQ = multiple-choice, SS = single-select; Zeroed = open-ended answers scored 0 for a safety red-line violation. † GPT-6 为 808 题（少 1 道选择题）。‡ DeepSeek V3.2 / V3：开放题用 V3.2、选择题用 V3 的混合结果。总分沿用原表，不由分项均值重算。总分列的进度条从 70 分起算，便于看出差距。带 <span class=\"hfb\">HF</span> 标记的模型可跳转到其 Hugging Face 模型页。", "empty": "没有匹配的模型", "h_scene": "主题榜", "s_scene": "三个业务场景各自排名。几乎所有模型都在 PG 个性化课程方案上得分最低。", "DA_t": "运动数据解读", "DA_d": "数据理解、指标解释与行动建议", "PG_t": "个性化课程方案", "PG_d": "目标适配、课程选择与方案编排", "QA_t": "专业知识问答", "QA_d": "知识准确、适用条件与表达质量", "top": "最高", "h_cap": "能力榜", "s_cap": "安全层、专业层、体验层并排对照。", "so_total": "按总分排序", "so_safe": "按安全层排序", "so_pro": "按专业层排序", "so_exp": "按体验层排序", "so_ms": "按多选排序", "so_oq": "按开放问答排序", "l_safe": "安全层", "l_pro": "专业层", "l_exp": "体验层", "l_ss": "单选", "l_ms": "多选", "l_oq": "开放问答", "h_type": "题型榜", "s_type": "单选、多选、开放问答并排对照。单选题全部满分，区分度主要来自开放问答。", "h_prog": "进步榜", "s_prog": "同系列两个版本之间的总分变化。这里的“进步”指版本间分数差，不代表时间速度，也不能单独归因于版本升级。", "n_prog": "Qwen 系列只连接 Max → Max；32B、27B、35B-A3B、235B-A22B、397B-A17B 属于不同规模或架构，不作为连续升级链。", "rankN": "增幅第 ", "pts": "分", "safeL": "安全", "h_about": "评测说明", "s_about": "评价框架、评分方式与数据口径。", "a_fw": "评价框架", "p_safe": "安全", "p_pro": "专业", "p_exp": "体验", "a_fw_l": "<li><b>DA 运动数据解读</b>：数据理解、指标解释与行动建议</li><li><b>PG 个性化课程方案</b>：目标适配、课程选择与方案编排</li><li><b>QA 专业知识问答</b>：知识准确、适用条件与表达质量</li>", "a_sc": "评分方式", "a_sc_l": "<li>开放题由三个 LLM judge 组成的加权评审团，依据专家编写的逐题 rubric 评分。</li><li>选择题（单选、多选）按标准答案精确匹配评分。</li><li>触犯安全红线的开放题记 0 分；总榜“红线置零”列为每个模型被置零的题数。</li>", "a_dn": "数据口径", "a_dn_l": "<li>分数为原表分数 × 100，保留 1 位小数。</li><li>本页为 809 题全量评测结果。公开的 80 题子集成绩见<a href=\"https://huggingface.co/datasets/KeepaceAI/MoveBench\">数据集页</a>，两者不可混用。</li><li>模型名称沿用原表写法。</li>", "a_sub": "参与评测", "a_sub_l": "<li>下载<a href=\"https://huggingface.co/datasets/KeepaceAI/MoveBench\">数据集</a>，用你的模型生成答案。</li><li>按<a href=\"https://github.com/Keep-Tech/MoveBench/blob/main/docs/SUBMISSION.md\">提交指南</a>提交答案文件，维护方评分审核后更新榜单。</li><li>引用：<a href=\"https://arxiv.org/abs/2610.26185\">MoveBench 论文</a>（arXiv 2610.26185）。</li>", "foot": "MoveBench Team × 首都体育学院 · 数据集与文档采用 CC BY 4.0 许可"}};
 let L='en';try{const s=localStorage.getItem('mb-lang');if(s==='zh'||s==='en')L=s;}catch(e){}
 const t=k=>T[L][k];
 const f=v=>(v*100).toFixed(1);
@@ -148,7 +148,7 @@ function renderTable(){
  document.querySelector('#tbl tbody').innerHTML=rows.length?rows.map(m=>`<tr class="r${m.rank}">
  <td class="l"><span class="rk">${m.rank}</span></td>
  <td class="l"><span class="mname">${esc(m.name)}${flag(m)}</span>${m.hf?`<a class="hfb" href="https://huggingface.co/${m.hf}" title="${m.hf}">HF</a>`:''}<span class="fam">${esc(m.family)}</span></td>
- <td><div class="tot"><b>${f(m.total)}</b><div class="mini"><i style="width:${m.total*100}%"></i></div></div></td>
+ <td><div class="tot"><b>${f(m.total)}</b><div class="mini" title="70–100"><i style="width:${Math.max(0,Math.min(100,(m.total*100-70)/30*100))}%"></i></div></div></td>
  <td>${f(m.oq)}</td><td>${f(m.mcq)}</td><td>${f(m.ss)}</td><td>${f(m.DA)}</td><td>${f(m.PG)}</td><td>${f(m.QA)}</td><td>${f(m.safe)}</td><td>${f(m.pro)}</td><td>${f(m.exp)}</td><td>${m.zero}</td></tr>`).join(''):`<tr><td colspan="13" class="l">${t('empty')}</td></tr>`;
 }
 function renderScenes(){