Security code review
结果只在同一分组内进行比较，因此用一种方式测试的模型绝不会与用另一种方式测试的模型同榜排名。
Find security vulnerabilities in real application code and fix them without breaking legitimate behavior
各排名分组使用独立的评分尺度。请只在同一分组内比较配置；EvalRank 绝不会把不同的系统压平到同一个榜单上。
跨基准共识
把该主题下已收录的基准合并成一份模型排名，权重与实际决策时一致。每个分数代表该模型成绩占最佳成绩的比例。
| 排名 | 模型 | 最佳占比 | 基准测试 |
|---|---|---|---|
| 1 | Kimi K3.1 Pro | 100.0% | 2 |
| 2 | MiMo-V2.6-Pro | 87.5% | 2 |
| 3 | GLM-5.3-Flash | 72.6% | 2 |
| 4 | Step 5 Preview | 64.3% | 2 |
| 5 | Beam 1.0 | 49.2% | 2 |
| 6 | Qwen3.8 2.4T A95B | 36.0% | 2 |
| 7 | MiniMax-M3 | 35.0% | 2 |
| 8 | Nemotron 3 Ultra 550B A55B | 23.6% | 2 |
智能体系统 · 智能体式
探索性——尚未产生排名冠军rg-security-code-review-agent-system-agentic-agent-system-v1Explorer 依据
CWE-Bench-AA (Artificial Analysis)
各来源不一致，分别展示，绝不取平均。观测于 ；有效至 。
| 排名 | 配置 | 能力 | 不确定性 | 家族 |
|---|---|---|---|---|
| 1 | DeepSeek V4.1 Flash 1017 (Max) cwe-bench-aa:f151fd0b-2139-40cf-a3ec-995e8cedc592config_b4568a44ba7d21a439edb6a56d1dd5bcae41e3cc0d1e7c6197f12cca591d4556 | 100% | 未知 | 1 |
| 2 | Grok 4.7 (Xhigh) cwe-bench-aa:59627802-1546-422c-9a65-65b61c022d36config_31b6c76643164be43b610600471ddb316446ebf1e6a9ad852853f1e470fb4199 | 98% | 未知 | 1 |
| 3 | GPT-6 Sol (Daybreak Blue, max) cwe-bench-aa:93e12518-6e04-4713-b52d-2c56dac49c63config_247334b28fd1aed37fa74acbd03ab41a7719f448751b986a3c5ca641902927bf | 90% | 未知 | 1 |
| 4 | GPT-6 Astra (Max) cwe-bench-aa:2f339a97-9a0d-499a-9cb5-e0db665bfa25config_0ff6c2e86d5fc3f8b315d1423793e587e2f52c6c7ae7e79ab9b3013eab9e95d2 | 86% | 未知 | 1 |
| 5 | MiMo-V2.6-Pro cwe-bench-aa:24d8fdfe-6241-424e-b7b7-1b6efb06e4fbconfig_78ea99edad6aab69cc66343588172e0229061ad766e52e4e070f86eeaa04fde9 | 86% | 未知 | 1 |
| 6 | Kimi K3.1 Pro cwe-bench-aa:665c67ca-09eb-48f0-a7a3-0096e1d5de51config_9cb2282d6a43fd40030819601e648892d6c1cd57e88bf0b09c92ed6c170870a2 | 80% | 未知 | 1 |
| 7 | Claude Opus 5.5 (Max, Default Fallback) cwe-bench-aa:2f3c4dc9-a450-4303-8697-0237257cf08fconfig_75aaaf8b916b7a66d01b00ce5506718dfaed135a097bd26331afc83e9ecf4ac1 | 75% | 未知 | 1 |
| 8 | Mistral Large 4 (High) cwe-bench-aa:ea5019ae-b0b7-49a5-80b3-0e584807af0fconfig_a03f5fba6bc3ec3837542b12cce927f15c1b58df7c0e7fab8ce2ea0eecff3f56 | 75% | 未知 | 1 |
| 9 | GPT-6 Luna (Max) cwe-bench-aa:6fb13851-40ce-4bda-ad3f-6b38e0c5daa7config_93cace1581371fb001fba9498b62acaf028700c5d458e2ef6afcc240ee4dd573 | 71% | 未知 | 1 |
| 10 | GLM-5.3 (Max) cwe-bench-aa:cd684ea4-b475-4269-b001-d469d06d8a7aconfig_3325cc082d7df8068a9cc64cf63782c025776594aa02f9958e0fd96e482ceb18 | 69% | 未知 | 1 |
| 11 | GLM-5.3-Flash cwe-bench-aa:19496b81-9f41-4214-a77a-1df803b3c5aeconfig_9f130b98f6a0318d0dcfd75f2dbad7ebf3128e51dca686e27609f3e179f9c8bb | 69% | 未知 | 1 |
| 12 | Gemini 3.8 Flash (High) cwe-bench-aa:8a999846-4c1d-4ce7-a8b7-1310a7166fd7config_354b3e670d3aa257f24e6d3e5c7badd5feb44c68dee0982822a4e0922dfad597 | 61% | 未知 | 1 |
| 13 | Step 5 Preview cwe-bench-aa:094ea3e3-ce41-419e-8bd1-723389663158config_67de31038e4fe36bc933bc68fc501555cd2288c2083f73088ac2ce3a04a5b7a2 | 61% | 未知 | 1 |
| 14 | Claude Fable 5.1 (Max, Default Fallback) cwe-bench-aa:3e87c73e-a257-495e-9730-367a66229811config_d8d4e1e7689af7c42f5688d542d5fb95f1a6b24903a29c9d49326ae537ee7b60 | 57% | 未知 | 1 |
| 15 | Beam 1.0 cwe-bench-aa:69468cce-7b5c-414a-b498-41df4c34df85config_b9982c75359bd843bfbd4ddd068638eb6f66a4c60c3d2d7d085e2d26230e50c1 | 53% | 未知 | 1 |
| 16 | MiniMax-M3 cwe-bench-aa:277f939a-985b-4b37-859d-b3eabc7c0b26config_907105a30b99df99fd3e6f83c042c3dbe0ca75fcc5bc8f41d816424fb62b8ecc | 35% | 未知 | 1 |
| 17 | Qwen3.8 2.4T A95B cwe-bench-aa:8df710d3-9dae-4498-9b4e-9818238e6f31config_5088d33050ecbde1e905be1ccaca5d512398e5eb11772e6788e5090614251df8 | 4% | 未知 | 1 |
| 18 | Nemotron 3 Ultra 550B A55B (Reasoning) cwe-bench-aa:5b52def2-ac9b-4465-ad80-91ea8079e253config_7290235f7802366ba63fb8e54a91694fab50b54aa4f5f2c3972a4730210d3683 | 2% | 未知 | 1 |
| 19 | Qwen3.8 27B (Xhigh) cwe-bench-aa:b01dee41-c62b-48ed-8d16-984adc405e5cconfig_2e3dbb3ed500173be12ba8b8d88ab36215a87834ec2f22cca0882bb419c47abd | 0% | 未知 | 1 |
Explorer 依据
CyberGym-E2E-AA (Artificial Analysis)
各来源不一致，分别展示，绝不取平均。观测于 ；有效至 。
| 排名 | 配置 | 能力 | 不确定性 | 家族 |
|---|---|---|---|---|
| 1 | Kimi K3.1 Pro cybergym-e2e-aa:665c67ca-09eb-48f0-a7a3-0096e1d5de51config_9137d86891c84a18fb0f16404dc0f74127b8336cf5c2961cb470147ea83128d6 | 100% | 未知 | 1 |
| 2 | GPT-6 Sol (Daybreak Blue, max) cybergym-e2e-aa:93e12518-6e04-4713-b52d-2c56dac49c63config_21cb86828f24e05190c62a8852d17592073156f79d4999b118dd1246e55266db | 98% | 未知 | 1 |
| 3 | Mistral Large 4 (High) cybergym-e2e-aa:ea5019ae-b0b7-49a5-80b3-0e584807af0fconfig_75c18b592f6fa928cd9ab15c0646f2db69adf8e33c9bf12f76f5c2a29634f0ad | 84% | 未知 | 1 |
| 4 | MiMo-V2.6-Pro cybergym-e2e-aa:24d8fdfe-6241-424e-b7b7-1b6efb06e4fbconfig_d2d43e8d5689df753e6392c341337b2e6dd4762d22e91dad49ea093d782eba56 | 83% | 未知 | 1 |
| 5 | GPT-6 Luna (Max) cybergym-e2e-aa:6fb13851-40ce-4bda-ad3f-6b38e0c5daa7config_33ac70fd3b281ee171b6c00d09b2ba45cef22c00bc0f392b52048434150a7684 | 82% | 未知 | 1 |
| 6 | DeepSeek V4.1 Flash 1017 (Max) cybergym-e2e-aa:f151fd0b-2139-40cf-a3ec-995e8cedc592config_59c1432804e10864949fdfc2b3185c602e32478e624c871aea6db35db7e02d2f | 81% | 未知 | 1 |
| 7 | Step 5 Preview cybergym-e2e-aa:094ea3e3-ce41-419e-8bd1-723389663158config_546c5989e471b276bcbe01827827e77780566ed9cea3c284d0bb7463c70bf6f4 | 79% | 未知 | 1 |
| 8 | GLM-5.3-Flash cybergym-e2e-aa:19496b81-9f41-4214-a77a-1df803b3c5aeconfig_060b05cfcfdb2dabc451bb994cfc7d89be873e6ddbddce28b666b70efd4919fb | 78% | 未知 | 1 |
| 9 | Grok 4.7 (Xhigh) cybergym-e2e-aa:59627802-1546-422c-9a65-65b61c022d36config_aaf52d8b90c5731ebf5d37fa62c258a20f387ee2e22835179b6a4079821ebd5a | 78% | 未知 | 1 |
| 10 | Beam 1.0 cybergym-e2e-aa:69468cce-7b5c-414a-b498-41df4c34df85config_0bd4d1c6ee6f21292a3de2af5416528d7201738dcee267761032f95d701c33e3 | 48% | 未知 | 1 |
| 11 | GLM-5.3 (Max) cybergym-e2e-aa:cd684ea4-b475-4269-b001-d469d06d8a7aconfig_59bb674568f8034b73ea573ab86aa3f9e0a0974eddea4995d75181e368108a10 | 31% | 未知 | 1 |
| 12 | MiniMax-M3 cybergym-e2e-aa:277f939a-985b-4b37-859d-b3eabc7c0b26config_5d8cf94c70c5afe5499a6664d6343bae4c693746b626fc7f98daf2d4004a6ae1 | 26% | 未知 | 1 |
| 13 | Nemotron 3 Ultra 550B A55B (Reasoning) cybergym-e2e-aa:5b52def2-ac9b-4465-ad80-91ea8079e253config_9f839de5800c532334b7301b3e2525ae7ef08ad1936d86c2755628a628d7497f | 11% | 未知 | 1 |
| 14 | Gemini 3.8 Flash (High) cybergym-e2e-aa:8a999846-4c1d-4ce7-a8b7-1310a7166fd7config_9da903d7339742b5ed052cab92fa9a59c1b4a2141673ce00df237005d4559eb6 | 9% | 未知 | 1 |
| 15 | Claude Fable 5.1 (Max, Default Fallback) cybergym-e2e-aa:3e87c73e-a257-495e-9730-367a66229811config_905f4d29a1510e42332187295d138b93d0f88844c7cf2fa6e8c69c0fb1a1ac24 | 1% | 未知 | 1 |
| 16 | Claude Opus 5.5 (Max, Default Fallback) cybergym-e2e-aa:2f3c4dc9-a450-4303-8697-0237257cf08fconfig_be9d5d4cfd8c60b5507cbcba14b62fde7659214df695b4d9a56d22a296fac86c | 1% | 未知 | 1 |
| 17 | GPT-6 Astra (Max) cybergym-e2e-aa:2f339a97-9a0d-499a-9cb5-e0db665bfa25config_1cd15a6088e4a68c6b73ebd0232967c15d229fa1ee90e5f603cf1d781d24c0f3 | 0% | 未知 | 1 |
| 18 | Qwen3.8 2.4T A95B cybergym-e2e-aa:8df710d3-9dae-4498-9b4e-9818238e6f31config_6b8c60882bfe0cc2b68ce67c24c8182a9dcdc88e285e3e66fd5a2b5782dc76f6 | 0% | 未知 | 1 |
| 19 | Qwen3.8 27B (Xhigh) cybergym-e2e-aa:b01dee41-c62b-48ed-8d16-984adc405e5cconfig_bc187f45ff6ce8d22223ae39007257df7b2a2670c4dcf933a103b3ad40c2558b | 0% | 未知 | 1 |
Explorer 依据
DeepsecBench-AA (Artificial Analysis)
各来源不一致，分别展示，绝不取平均。观测于 ；有效至 。
| 排名 | 配置 | 能力 | 不确定性 | 家族 |
|---|---|---|---|---|
| 1 | GPT-6 Sol (Daybreak Blue, max) deepsecbench-aa:93e12518-6e04-4713-b52d-2c56dac49c63config_6b3c5e4139967db796ac1b932c7b35fbc294917d5911fd4ab09c2ff99dbaf9f6 | 100% | 未知 | 1 |
| 2 | GPT-6 Astra (Max) deepsecbench-aa:2f339a97-9a0d-499a-9cb5-e0db665bfa25config_4d9b933720cbb942b3ce8e62a6339d2fc1a934cc2886c88c83ce213d2932b597 | 76% | 未知 | 1 |
| 3 | Kimi K3.1 Pro deepsecbench-aa:665c67ca-09eb-48f0-a7a3-0096e1d5de51config_484e33155b0d79051491cc0de893c00de697a4e9451dd61179f53e66dfc4df4e | 76% | 未知 | 1 |
| 4 | Grok 4.7 (Xhigh) deepsecbench-aa:59627802-1546-422c-9a65-65b61c022d36config_831bccb5b1f4d3c61244297d49ac1330bfd97bab0ddf73b410b12c70ac24cdca | 55% | 未知 | 1 |
| 5 | Claude Opus 5.5 (Max, Default Fallback) deepsecbench-aa:2f3c4dc9-a450-4303-8697-0237257cf08fconfig_93873bfa5f379c9cf84ddafd82635b56167f772bbcfcb317e684b8c87ea70f3d | 55% | 未知 | 1 |
| 6 | MiMo-V2.6-Pro deepsecbench-aa:24d8fdfe-6241-424e-b7b7-1b6efb06e4fbconfig_d4d043bf707aaab71397052547a416529a08d6e16b739c8ea24087e6bd654e35 | 54% | 未知 | 1 |
| 7 | DeepSeek V4.1 Flash 1017 (Max) deepsecbench-aa:f151fd0b-2139-40cf-a3ec-995e8cedc592config_a6aac67405fb7f3a3f1629b02f60f9c31dec0cdd6706af76154ceb810884250b | 51% | 未知 | 1 |
| 8 | GLM-5.3 (Max) deepsecbench-aa:cd684ea4-b475-4269-b001-d469d06d8a7aconfig_0b2300dc53b9019c768961da18967120d4c054ec3c438c6fe19e9264f6aae2d3 | 49% | 未知 | 1 |
| 9 | GPT-6 Luna (Max) deepsecbench-aa:6fb13851-40ce-4bda-ad3f-6b38e0c5daa7config_40f58707d5621e7b3fd5a930856442c0b156820badbdea42a81d6c88a270a9a9 | 48% | 未知 | 1 |
| 10 | Claude Fable 5.1 (Max, Default Fallback) deepsecbench-aa:3e87c73e-a257-495e-9730-367a66229811config_adb39dd35b0fab85cac9cb9dec225f81ff4a0f47c2772e8d346802bbad63088f | 48% | 未知 | 1 |
| 11 | GLM-5.3-Flash deepsecbench-aa:19496b81-9f41-4214-a77a-1df803b3c5aeconfig_4a7373829e2784f35fc615103f058cc9ef795a9f49f28820ee28bac22d94ac2f | 40% | 未知 | 1 |
| 12 | Step 5 Preview deepsecbench-aa:094ea3e3-ce41-419e-8bd1-723389663158config_2ea19caff21ede7ec45f62bccbc156559dadfc85677a66b81e389e32099b5df0 | 31% | 未知 | 1 |
| 13 | Mistral Large 4 (High) deepsecbench-aa:ea5019ae-b0b7-49a5-80b3-0e584807af0fconfig_37c2736423c488bc368155e07d7ac2545d7f1345c77d6c0550b37975aede9c4a | 30% | 未知 | 1 |
| 14 | Gemini 3.8 Flash (High) deepsecbench-aa:8a999846-4c1d-4ce7-a8b7-1310a7166fd7config_ea7eac4b18f0b6869b68409457a9647c12b4868d8c59b627ace49396a5fd8c4c | 29% | 未知 | 1 |
| 15 | Qwen3.8 27B (Xhigh) deepsecbench-aa:b01dee41-c62b-48ed-8d16-984adc405e5cconfig_97991e5d00071c023c2aeca1aa27590564ae7ecdf572235d678b2e30dbe17404 | 22% | 未知 | 1 |
| 16 | Qwen3.8 2.4T A95B deepsecbench-aa:8df710d3-9dae-4498-9b4e-9818238e6f31config_28ef97e53dadeaffd8bf2b6d398cecfefec5d52628dd6f48468da2d819d61fa8 | 18% | 未知 | 1 |
| 17 | Beam 1.0 deepsecbench-aa:69468cce-7b5c-414a-b498-41df4c34df85config_22555ebb3170c6e7adf18b55bf668a5bd52d6b59bc33c691918eb821ff8d1435 | 12% | 未知 | 1 |
| 18 | MiniMax-M3 deepsecbench-aa:277f939a-985b-4b37-859d-b3eabc7c0b26config_714a4e54aa58e532b07b25e41b4c0cd4cf91abb8c681f333ac412a1acd87d6a3 | 0% | 未知 | 1 |
| 19 | Nemotron 3 Ultra 550B A55B (Reasoning) deepsecbench-aa:5b52def2-ac9b-4465-ad80-91ea8079e253config_4051bb5bdcf64e85605487923ef99a98cdb7243c56b3ca171d182dcc45eec753 | 0% | 未知 | 1 |
Explorer 依据
DeepsecBench-AA refusal rate (Artificial Analysis)
各来源不一致，分别展示，绝不取平均。观测于 ；有效至 。
对比该分组中的配置
独立依据不足，无法形成头部梯队
- 校准尚未通过验证
- 配置重叠度低于要求的门槛
- 独立基准族的覆盖度低于要求的门槛