OTIS Mock AIME 2024-2025
Often cited- What does it measure?
- Tests math problem solving with mock exams written by students of the OTIS olympiad-training program to imitate the AIME.
- Who evaluates it, and how?
- 45 problems, 15 each from Mock AIME 2024, 2025 (I), and 2025 (II). Every answer is an integer from 0 to 999; the AI's answer is extracted and checked against the answer key.
- Reading the score
- Scores run from 0 to 100%. Higher is better. It is harder than MATH Level 5 and easier than FrontierMath.
Top score100.0%
Models tested80
Scoring basisHigher is better
Last updated2026.11.06
Model release date
Can it solve problems that stump experts across fields?Humanity's Last Exam
Claude Opus 5.5🥇
Claude Opus 5.5🥇Humanity's Last Exam
Claude Opus 5.5🥇Can it solve research-level physics problems?CritPt
Gemini 4 Carbon🥇
Gemini 4 Carbon🥇CritPt
Gemini 4 Carbon🥇Which AI gives the math answers people prefer most?Arena Math
Gemini 4 Argon🥇
Gemini 4 Argon🥇Arena Math
Gemini 4 Argon🥇
- 🥇AnthropicClaude Fable 5.1Reasoning effort: Max100.0%
- 🥇GoogleGemini 4 ArgonReasoning effort: High100.0%
- 🥇AnthropicClaude Opus 5.5Reasoning effort: Max100.0%
- 🥇AnthropicClaude Fable 5Reasoning effort: High100.0%
- 🥇OpenAIGPT-6.1 TerraReasoning effort: Max100.0%
- 🥇OpenAIGPT-6 SolReasoning effort: Max100.0%
- 🥇OpenAIGPT-6.2 AstraReasoning effort: Max100.0%
- 🥇OpenAIGPT-6.1 SolReasoning effort: Max100.0%
- 🥇GoogleGemini 4 CarbonReasoning effort: High100.0%
- 🥇OpenAIGPT-6 AstraReasoning effort: Max100.0%
- 🥇OpenAIGPT-5.6 SolReasoning effort: Max100.0%
- 🥇AnthropicClaude Sonnet 5.5Reasoning effort: Max100.0%
- #13AnthropicClaude Fable 5.2Reasoning effort: Max99.7%
- #13OpenAIGPT-5.6 TerraReasoning effort: Max99.7%
- #15DeepSeekDeepSeek V4.1 Flash 1017Reasoning effort: Max99.4%
- #16SpaceXSIGrok 4.6Reasoning effort: Extra High99.2%
- #16MetaMuse Spark 1.3Reasoning effort: Extra High99.2%
- #18OpenAIGPT-6 LunaReasoning effort: Max98.9%
- #18AnthropicClaude Haiku 5.5Reasoning effort: Extra High98.9%
- #18GoogleGemini 3.8 FlashReasoning effort: High98.9%
- #18AnthropicClaude Opus 5Reasoning effort: Max98.9%
- #22DeepSeekDeepSeek V4 ProReasoning effort: Max98.6%
- #23DeepSeekDeepSeek V4.1 FlashReasoning effort: Max98.3%
- #23OpenAIGPT-5.6 LunaReasoning effort: Max98.3%
- #23AnthropicClaude Opus 4.8Reasoning effort: Max98.3%
- #23Mistral AIMistral Large 4Reasoning effort: High98.3%
- #27SpaceXSIGrok 4.7Reasoning effort: Extra High98.1%
- #28AnthropicClaude Opus 4.7Reasoning effort: Extra High97.8%
- #29SpaceXSIGrok 4.5Reasoning effort: High97.8%
- #29OpenAIGPT-5.4Reasoning effort: High97.8%
- #31Moonshot AIKimi K3Reasoning effort: Max97.2%
- #31GoogleGemini 3.7 FlashReasoning effort: High97.2%
- #33Moonshot AIKimi K2.696.1%
- #34GoogleGemini 3.1 Pro95.6%
- #35GoogleGemini 3 FlashReasoning effort: High95.6%
- #35GoogleGemini 3.5 FlashReasoning effort: High95.6%
- #35AlibabaQwen3.7 Max95.6%
- #35Moonshot AIKimi K2.7 Code95.6%
- #39AnthropicClaude Sonnet 5Reasoning effort: Extra High94.7%
- #40AnthropicClaude Opus 4.6Reasoning effort: Budget 64K94.4%
- #40DeepSeekDeepSeek V4 FlashReasoning effort: Max94.4%
- #42GoogleGemini 3.6 FlashReasoning effort: High94.2%
- #43Z.aiGLM 5.3 FlashReasoning effort: Max93.9%
- #44AlibabaQwen3.6 Plus93.3%
- #44AlibabaQwen3.7 Plus93.3%
- #44SpaceXSIGrok 4.3Reasoning effort: High93.3%
- #44Z.aiGLM-5.193.3%
- #48SpaceXSIGrok 4.20 (Reasoning)92.2%
- #49OpenAIGPT-5Reasoning effort: High91.4%
- #50Z.aiGLM 5.3Reasoning effort: Max91.1%
- #50AlibabaQwen3.6 Max91.1%
- #52MetaMuse Spark88.9%
- #53OpenAIGPT OSS 120BReasoning effort: High88.9%
- #53AlibabaQwen3.5 397B A17B88.9%
- #53OpenAIGPT-5.4 MiniReasoning effort: Extra High88.9%
- #56OpenAIGPT-5.4 NanoReasoning effort: High87.8%
- #57NVIDIANemotron 3 Ultra86.7%
- #57AlibabaQwen3.5 Plus86.7%
- #57AlibabaQwen3.6 35B A3B86.7%
- #57OpenAIGPT-5 MiniReasoning effort: High86.7%
- #61Z.aiGLM 5.2Reasoning effort: Max86.4%
- #62AnthropicClaude Opus 4.5Reasoning effort: Budget 32K86.1%
- #63AnthropicClaude Sonnet 4.6Reasoning effort: Budget 32K85.8%
- #64GoogleGemini 2.5 Pro84.7%
- #65OpenAIGPT-5.5Reasoning effort: Low84.4%
- #65AlibabaQwen3.6 Flash84.4%
- #67OpenAIGPT-5 NanoReasoning effort: High81.1%
- #68GoogleGemini 3.1 Flash LiteReasoning effort: High80.0%
- #68Z.aiGLM-580.0%
- #70AnthropicClaude Sonnet 4.5Reasoning effort: Budget 59K77.8%
- #71GoogleGemma 4 31BReasoning effort: Minimal73.3%
- #72GoogleGemini 3.5 Flash-LiteReasoning effort: High71.1%
- #72AnthropicClaude Sonnet 4Reasoning effort: Budget 32K71.1%
- #72MiniMaxMiniMax M371.1%
- #75AnthropicClaude Opus 4.1Reasoning effort: Budget 27K68.9%
- #76AnthropicClaude Haiku 4.5Reasoning effort: Budget 32K66.7%
- #77AnthropicClaude Opus 4Reasoning effort: Budget 27K64.4%
- #78OpenAIGPT-4.138.3%
- #79MetaLlama 4 Maverick20.6%
- #80MetaLlama 4 Scout7.8%
Top 27 near 100.0% · hard to separate the best · Half of models ≤ 94.4%
The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.