benchgap
Calibration

Gert Labs → Agents' Last Exam

Agents' Last Exam is estimated from Gert Labs with a Hill curve fitted on 16 models measured on both: y = 0.2956 + (0.8314 − 0.2956)·x^6.00 / (0.86661^6.00 + x^6.00), R² = 0.58, cross-validated error 9.7 pp. It is used for 90 estimates.

Estimated modelGert LabsAgents' Last ExamSource
GPT-6.2 Astra90.0%59.4%estimated ± 9.7 pp, medium confidence
Claude Fable 5.288.2%57.8%estimated ± 9.7 pp, medium confidence
GPT-6.2 Sol86.8%56.5%estimated ± 9.7 pp, medium confidence
Gemini 4 Neon83.8%53.7%estimated ± 9.7 pp, medium confidence
Gemini 4 Hydrogen79.1%49.2%estimated ± 9.7 pp, medium confidence
Claude Opus 577.4%47.6%estimated ± 9.7 pp, medium confidence
Claude Fable 5.177.0%47.3%estimated ± 9.7 pp, medium confidence
Claude Fable 572.7%43.4%estimated ± 9.7 pp, medium confidence
Grok 4.771.5%42.4%estimated ± 9.7 pp, medium confidence
Qwen4 Max70.2%41.3%estimated ± 9.7 pp, medium confidence
Claude Sonnet 5.569.9%41.1%estimated ± 9.7 pp, medium confidence
Grok 4.669.2%40.6%estimated ± 9.7 pp, medium confidence
Gemini 3.8 Flash67.0%39.0%estimated ± 9.7 pp, medium confidence
Muse Wonder65.9%38.3%estimated ± 9.7 pp, medium confidence
Kimi K363.1%36.5%estimated ± 9.7 pp, medium confidence
Gemini 3.6 Flash59.5%34.6%estimated ± 9.7 pp, medium confidence
Claude Haiku 5.558.6%34.2%estimated ± 9.7 pp, medium confidence
Ember-158.0%34.0%estimated ± 9.7 pp, medium confidence
Muse Spark 1.357.0%33.6%estimated ± 9.7 pp, medium confidence
Muse Ember56.1%33.2%estimated ± 9.7 pp, medium confidence
Claude Opus 4.855.6%33.0%estimated ± 9.7 pp, medium confidence
GPT-5.3 Codex52.0%31.9%estimated ± 9.7 pp, medium confidence
Mercury 350.6%31.6%estimated ± 9.7 pp, medium confidence
GLM-5.249.1%31.3%estimated ± 9.7 pp, medium confidence
Inkling 2 Small48.6%31.2%estimated ± 9.7 pp, medium confidence
Gemini 3.5 Flash48.6%31.2%estimated ± 9.7 pp, medium confidence
GPT-5.247.7%31.0%estimated ± 9.7 pp, medium confidence
Qwen 3.6 Max (preview)47.5%31.0%estimated ± 9.7 pp, medium confidence
Gemini 3 Pro47.4%31.0%estimated ± 9.7 pp, medium confidence
Grok 4.547.2%30.9%estimated ± 9.7 pp, medium confidence
Claude Sonnet 547.0%30.9%estimated ± 9.7 pp, medium confidence
Kimi K2.647.0%30.9%estimated ± 9.7 pp, medium confidence
Claude Opus 4.546.0%30.7%estimated ± 9.7 pp, medium confidence
Gemini 3.1 Pro45.7%30.7%estimated ± 9.7 pp, medium confidence
Muse Spark 1.245.2%30.6%estimated ± 9.7 pp, medium confidence
Qwen3.7 Max44.8%30.6%estimated ± 9.7 pp, medium confidence
MiMo-V2.5-Pro44.2%30.5%estimated ± 9.7 pp, medium confidence
Muse Spark Open43.6%30.4%estimated ± 9.7 pp, medium confidence
Kimi K2.7 Code40.7%30.1%estimated ± 9.7 pp, medium confidence
GPT-5.139.9%30.1%estimated ± 9.7 pp, medium confidence
Mistral Large 439.8%30.1%estimated ± 9.7 pp, medium confidence
Hy3 Preview39.3%30.0%estimated ± 9.7 pp, medium confidence
Beam 1.0 501B-A23B36.9%29.9%estimated ± 9.7 pp, medium confidence
Kimi K2.535.5%29.8%estimated ± 9.7 pp, medium confidence
Qwen3.7 Plus35.3%29.8%estimated ± 9.7 pp, medium confidence
Claude Sonnet 4.534.9%29.8%estimated ± 9.7 pp, medium confidence
Gemini 3 Flash34.2%29.8%estimated ± 9.7 pp, low confidence
Muse Glimmer 30B32.7%29.7%estimated ± 9.7 pp, low confidence
Gemini 2.5 Pro32.6%29.7%estimated ± 9.7 pp, low confidence
Qwen3.6-27B32.4%29.7%estimated ± 9.7 pp, low confidence
Grok 431.6%29.7%estimated ± 9.7 pp, low confidence
Qwen3.6 Plus31.4%29.7%estimated ± 9.7 pp, low confidence
Nemotron 3.5 Super VL 120B A12B30.9%29.7%estimated ± 9.7 pp, low confidence
Grok 4.20 Multi-agent30.2%29.7%estimated ± 9.7 pp, low confidence
MiMo-V2.529.9%29.7%estimated ± 9.7 pp, low confidence
MiniMax M329.8%29.6%estimated ± 9.7 pp, low confidence
Grok 4.1 Fast29.3%29.6%estimated ± 9.7 pp, low confidence
Qwen3.5 397B29.2%29.6%estimated ± 9.7 pp, low confidence
Qwen3.5-27B28.9%29.6%estimated ± 9.7 pp, low confidence
MiniMax M2.728.8%29.6%estimated ± 9.7 pp, low confidence
GPT-5.1-Codex28.1%29.6%estimated ± 9.7 pp, low confidence
Claude 4 Sonnet27.7%29.6%estimated ± 9.7 pp, low confidence
Gemma 4 31B27.7%29.6%estimated ± 9.7 pp, low confidence
GLM-527.7%29.6%estimated ± 9.7 pp, low confidence
GPT-5.2-Codex27.3%29.6%estimated ± 9.7 pp, low confidence
Nemotron 3 Ultra27.3%29.6%estimated ± 9.7 pp, low confidence
Grok Build 0.127.2%29.6%estimated ± 9.7 pp, low confidence
Inkling26.6%29.6%estimated ± 9.7 pp, low confidence
Hy326.3%29.6%estimated ± 9.7 pp, low confidence
Gemini 3.1 Flash-Lite25.7%29.6%estimated ± 9.7 pp, low confidence
Trinity-Large-Thinking25.4%29.6%estimated ± 9.7 pp, low confidence
MiMo-V2-Pro24.7%29.6%estimated ± 9.7 pp, low confidence
Qwen3.7 Flash24.6%29.6%estimated ± 9.7 pp, low confidence
GLM-4.723.8%29.6%estimated ± 9.7 pp, low confidence
Gemini 3.5 Flash-Lite23.8%29.6%estimated ± 9.7 pp, low confidence
GLM-5V-Turbo23.5%29.6%estimated ± 9.7 pp, low confidence
Grok 4.2022.3%29.6%estimated ± 9.7 pp, low confidence
Qwen3.6-35B-A3B21.5%29.6%estimated ± 9.7 pp, low confidence
DeepSeek V3.221.1%29.6%estimated ± 9.7 pp, low confidence
GPT-OSS 120B21.1%29.6%estimated ± 9.7 pp, low confidence
Grok 4.321.0%29.6%estimated ± 9.7 pp, low confidence
Mistral Medium 3.5 128B20.7%29.6%estimated ± 9.7 pp, low confidence
Laguna M.120.3%29.6%estimated ± 9.7 pp, low confidence
Step 3.7 Flash20.3%29.6%estimated ± 9.7 pp, low confidence
Qwen3.5-35B-A3B20.2%29.6%estimated ± 9.7 pp, low confidence
Mercury 2.520.0%29.6%estimated ± 9.7 pp, low confidence
Gemma 4 26B A4B19.6%29.6%estimated ± 9.7 pp, low confidence
GPT-4.118.9%29.6%estimated ± 9.7 pp, low confidence
Laguna S 2.117.5%29.6%estimated ± 9.7 pp, low confidence
Laguna XS 2.117.0%29.6%estimated ± 9.7 pp, low confidence

Fitted on

ModelGert LabsAgents' Last Exam
DeepSeek V4.1 Flash 101760.1%35.4%
DeepSeek V4 Flash 073146.8%25.2%
DeepSeek V4 Pro52.7%25.7%
Gemini 3.7 Flash61.4%26.3%
Gemini 4 Argon80.6%39.5%
GLM-5.355.0%28.5%
GLM-5.3-Flash49.6%26.3%
GPT-6.1 Terra83.2%60.0%
GPT-6 Astra92.0%59.3%
GPT-6 Sol79.0%56.4%
Kimi K3.1 Pro65.8%43.2%
MiMo-V2.6-Flash53.5%27.6%
MiMo-V2.6-Pro59.5%31.6%
MiniMax M3.1 Flash42.5%29.6%
Qwen3.8-27B34.5%42.9%
Qwen3.8 Max55.7%52.4%