OpenCompass LLM

Indexed results from OpenCompass LLM. Each ranking retains its own metric and evaluation conditions.

Math_Competition

openCompass · OpenCompass LLM · Math_Competition · percent · Higher is better

  1. MoonshotAI: Kimi K2.6moonshotai68.6%
  2. DeepSeek: DeepSeek V4 Flash 0423deepseek65.7%
  3. Tencent: Hy3 previewtencent63.5%
  4. DeepSeek: DeepSeek V4 Pro 0423deepseek62.5%
  5. MiniMax: MiniMax M3minimax51.6%

Knowledge_Social_Science

openCompass · OpenCompass LLM · Knowledge_Social_Science · percent · Higher is better

  1. DeepSeek: DeepSeek V4 Flash 0423deepseek87.1%
  2. DeepSeek: DeepSeek V4 Pro 0423deepseek87.1%
  3. Tencent: Hy3 previewtencent86.3%
  4. MoonshotAI: Kimi K2.6moonshotai85.8%
  5. MiniMax: MiniMax M3minimax82.5%

Common_Reasoning

openCompass · OpenCompass LLM · Common_Reasoning · percent · Higher is better

  1. MoonshotAI: Kimi K2.6moonshotai78.1%
  2. DeepSeek: DeepSeek V4 Flash 0423deepseek73%
  3. Tencent: Hy3 previewtencent72.2%
  4. MiniMax: MiniMax M3minimax69.9%
  5. DeepSeek: DeepSeek V4 Pro 0423deepseek69.6%

Knowledge_CommonSense

openCompass · OpenCompass LLM · Knowledge_CommonSense · percent · Higher is better

  1. DeepSeek: DeepSeek V4 Pro 0423deepseek92.5%
  2. MoonshotAI: Kimi K2.6moonshotai89.6%
  3. MiniMax: MiniMax M3minimax84.2%
  4. Tencent: Hy3 previewtencent84.2%
  5. DeepSeek: DeepSeek V4 Flash 0423deepseek77.9%

Knowledge_Humanity

openCompass · OpenCompass LLM · Knowledge_Humanity · percent · Higher is better

  1. DeepSeek: DeepSeek V4 Pro 0423deepseek95%
  2. MoonshotAI: Kimi K2.6moonshotai92.5%
  3. DeepSeek: DeepSeek V4 Flash 0423deepseek90%
  4. MiniMax: MiniMax M3minimax87.1%
  5. Tencent: Hy3 previewtencent83.8%

Math_College

openCompass · OpenCompass LLM · Math_College · percent · Higher is better

  1. MoonshotAI: Kimi K2.6moonshotai82.7%
  2. Tencent: Hy3 previewtencent81%
  3. DeepSeek: DeepSeek V4 Pro 0423deepseek79.9%
  4. DeepSeek: DeepSeek V4 Flash 0423deepseek75%
  5. MiniMax: MiniMax M3minimax74.3%

AIME2025

openCompass · OpenCompass LLM · AIME2025 · percent · Higher is better

  1. deepseek-ai/DeepSeek-V3.2-Specialedeepseek96%
  2. Z.ai: GLM 5z-ai95.8%
  3. StepFun: Step 3.5 Flashstepfun95.7%
  4. Z.ai: GLM 4.7z-ai95.4%
  5. MoonshotAI: Kimi K2 Thinkingmoonshotai94.1%
  6. OpenAI: gpt-oss-120bopenai93.4%
  7. DeepSeek: DeepSeek V3.2deepseek93%
  8. Qwen: Qwen3.5 397B A17Bqwen92.3%
  9. Qwen: Qwen3 235B A22B Thinking 2507qwen90.9%
  10. Z.ai: GLM 4.6z-ai90.3%
  11. DeepSeek: R1 0528deepseek89%
  12. Qwen: Qwen3 Next 80B A3B Thinkingqwen89%

Knowledge

openCompass · OpenCompass LLM · Knowledge · percent · Higher is better

  1. DeepSeek: DeepSeek V4 Pro 0423deepseek92.5%
  2. MoonshotAI: Kimi K2.6moonshotai89.8%
  3. MiniMax: MiniMax M3minimax86.6%
  4. Tencent: Hy3 previewtencent85.6%
  5. DeepSeek: DeepSeek V4 Flash 0423deepseek84.5%

Comprehensive_Coding_Ability

openCompass · OpenCompass LLM · Comprehensive_Coding_Ability · percent · Higher is better

  1. DeepSeek: DeepSeek V4 Pro 0423deepseek91%
  2. Tencent: Hy3 previewtencent82%
  3. DeepSeek: DeepSeek V4 Flash 0423deepseek79%
  4. MoonshotAI: Kimi K2.6moonshotai74.5%
  5. MiniMax: MiniMax M3minimax72.5%

Reasoning

openCompass · OpenCompass LLM · Reasoning · percent · Higher is better

  1. MoonshotAI: Kimi K2.6moonshotai62.9%
  2. DeepSeek: DeepSeek V4 Flash 0423deepseek61%
  3. DeepSeek: DeepSeek V4 Pro 0423deepseek55.5%
  4. Tencent: Hy3 previewtencent54.6%
  5. MiniMax: MiniMax M3minimax54%

Knowledge_Engineering

openCompass · OpenCompass LLM · Knowledge_Engineering · percent · Higher is better

  1. DeepSeek: DeepSeek V4 Pro 0423deepseek90.8%
  2. MoonshotAI: Kimi K2.6moonshotai90.8%
  3. MiniMax: MiniMax M3minimax88.3%
  4. Tencent: Hy3 previewtencent85%
  5. DeepSeek: DeepSeek V4 Flash 0423deepseek83.3%

Average

openCompass · OpenCompass LLM · Average · percent · Higher is better

  1. DeepSeek: DeepSeek V4 Pro 0423deepseek77.5%
  2. MoonshotAI: Kimi K2.6moonshotai75.7%
  3. DeepSeek: DeepSeek V4 Flash 0423deepseek73.7%
  4. Tencent: Hy3 previewtencent73.6%
  5. MiniMax: MiniMax M3minimax69%

Models

Loading
EVIDENCE / COMPARISON

Benchmarks

Explore every benchmark

Loading

These charts contain model-linked results. The charts above also include unlinked source profiles. Compare scores within each evaluation. Model quality, usage, transparency and system performance measure different things.

Install TensorTrace