diff --git a/bash/run.py b/bash/run.py
index 0b229ca..f41870f 100644
--- a/bash/run.py
+++ b/bash/run.py
@@ -148,22 +148,17 @@ SUITES = {
         'single': ['openai_mrcr', 'longbench_v2', 'bfcl_v3', 'mmlu', 'cmmlu', 'bbh', 'simple_qa'],
         'agent': ['tau2_bench', 'general_fc'],
     },
-    # 腾讯/官方入库标准对比套件：覆盖各厂商公开对比表里的全部 benchmark
-    # 注：swe_bench_*/terminal_bench_v2/browsecomp/mcp_atlas 需要额外镜像/工具
+    # 腾讯/官方入库标准对比套件：取各厂商公开对比表的并集，约 11 个 benchmark
+    # 注：swe_bench_verified/terminal_bench_v2/browsecomp/tau2_bench 需要额外镜像/工具
     'official': {
-        'multi': [
-            'aime24', 'aime25', 'aime26', 'hmmt26',
-            'imo_answerbench', 'live_code_bench', 'humaneval', 'gpqa_diamond',
-        ],
+        'multi': [],
         'single': [
-            'bigcodebench', 'bfcl_v3', 'competition_math', 'gsm8k', 'hle', 'super_gpqa',
-            'arc', 'bbh', 'cmmlu', 'drop', 'hellaswag', 'mmlu', 'mmlu_pro',
-            'simple_qa', 'trivia_qa', 'winogrande', 'openai_mrcr', 'longbench_v2',
-            'swe_bench_verified', 'swe_bench_pro', 'swe_bench_multilingual_agentic',
-            'terminal_bench_v2',
+            'aime25', 'aime26',
+            'hle', 'mmlu_pro', 'gpqa_diamond',
+            'longbench_v2', 'live_code_bench', 'swe_bench_verified',
         ],
         'agent': [
-            'tau2_bench', 'tau2_bench_retail', 'browsecomp', 'mcp_atlas', 'general_fc',
+            'tau2_bench', 'terminal_bench_v2', 'browsecomp',
         ],
     },
 }
diff --git a/myread.md b/myread.md
index 951561d..be5a666 100644
--- a/myread.md
+++ b/myread.md
@@ -409,7 +409,7 @@ python bash/run.py --suite full --thinking --folder-name my_exp_v1 --limit none
 | `group1` | 代码 + 数学/推理 | 约 **22h** | 多机组 1 |
 | `group2` | 重知识 | 约 **26h** | 多机组 2 |
 | `group3` | 长文本 + Agent + 知识 | 约 **27h** | 多机组 3 |
-| `official` | 官方对比表全部 benchmark | 视配置而定 | 覆盖 Kimi/GLM/DS 等公开对比表里的所有任务 |
+| `official` | `aime25`, `aime26`, `hle`, `mmlu_pro`, `gpqa_diamond`, `longbench_v2`, `live_code_bench`, `swe_bench_verified`, `tau2_bench`, `terminal_bench_v2`, `browsecomp` | 视配置而定 | 腾讯入库标准对比表的并集，约 11 个 benchmark；其中 swe/terminal/browsecomp/tau2 需额外镜像/工具 |
 
 > **分组时间说明**
 >