HELM Capabilities

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

HELM GPQA

helm · HELM Capabilities · chain_of_thought_correctness · percent

  1. gemini-3-pro-previewgoogle80.269%
  2. Qwen3-235B-A22B-Instruct-2507-FP8qwen72.646%
  3. OpenAI: gpt-oss-120bopenai68.386%
  4. Claude Opus 4anthropic66.592%
  5. DeepSeek: R1 0528deepseek66.592%
  6. Writer: Palmyra X5writer66.143%
  7. MoonshotAI: Kimi K2 0711moonshotai65.247%
  8. Llama-4-Maverick-17B-128E-Instruct-FP8meta-llama65.022%
  9. Claude Sonnet 4anthropic64.35%
  10. Qwen: Qwen3 Next 80B A3B Thinkingqwen63.004%
  11. OpenAI: gpt-oss-20bopenai59.417%
  12. GLM-4.5-Air-FP8z-ai59.417%

HELM IFEval

helm · HELM Capabilities · ifeval_strict_accuracy · percent

  1. Claude Opus 4anthropic91.774%
  2. Llama-4-Maverick-17B-128E-Instruct-FP8meta-llama90.758%
  3. gemini-3-pro-previewgoogle87.646%
  4. MoonshotAI: Kimi K2 0711moonshotai85.028%
  5. Claude Sonnet 4anthropic83.888%
  6. OpenAI: gpt-oss-120bopenai83.611%
  7. Qwen3-235B-A22B-Instruct-2507-FP8qwen83.518%
  8. DeepSeek: DeepSeek V3deepseek83.241%
  9. Writer: Palmyra X5writer82.286%
  10. OpenAI: GPT-4o (2024-11-20)openai81.731%
  11. Amazon: Nova Pro 1.0amazon81.547%
  12. GLM-4.5-Air-FP8z-ai81.177%

HELM MMLU-Pro

helm · HELM Capabilities · chain_of_thought_correctness · percent

  1. gemini-3-pro-previewgoogle90.3%
  2. Claude Opus 4anthropic85.9%
  3. Qwen3-235B-A22B-Instruct-2507-FP8qwen84.4%
  4. Claude Sonnet 4anthropic84.3%
  5. MoonshotAI: Kimi K2 0711moonshotai81.9%
  6. Llama-4-Maverick-17B-128E-Instruct-FP8meta-llama81%
  7. Writer: Palmyra X5writer80.4%
  8. OpenAI: gpt-oss-120bopenai79.5%
  9. DeepSeek: R1 0528deepseek79.3%
  10. Qwen: Qwen3 Next 80B A3B Thinkingqwen78.6%
  11. GLM-4.5-Air-FP8z-ai76.2%
  12. OpenAI: gpt-oss-20bopenai74%

HELM Omni-MATH

helm · HELM Capabilities · omni_math_accuracy · percent

  1. Qwen3-235B-A22B-Instruct-2507-FP8qwen71.833%
  2. OpenAI: gpt-oss-120bopenai68.833%
  3. MoonshotAI: Kimi K2 0711moonshotai65.4%
  4. OpenAI: gpt-oss-20bopenai56.467%
  5. gemini-3-pro-previewgoogle55.55%
  6. Claude Sonnet 4anthropic51.25%
  7. Claude Opus 4anthropic51.117%
  8. Google: Gemini 2.5 Flash Litegoogle47.967%
  9. Qwen: Qwen3 Next 80B A3B Thinkingqwen46.683%
  10. DeepSeek: R1 0528deepseek42.4%
  11. Llama-4-Maverick-17B-128E-Instruct-FP8meta-llama42.233%
  12. Writer: Palmyra X5writer41.45%

HELM WildBench

helm · HELM Capabilities · wildbench_score_rescaled · percent

  1. Qwen3-235B-A22B-Instruct-2507-FP8qwen86.626%
  2. MoonshotAI: Kimi K2 0711moonshotai86.187%
  3. gemini-3-pro-previewgoogle85.872%
  4. OpenAI: gpt-oss-120bopenai84.494%
  5. Claude Opus 4anthropic83.3%
  6. DeepSeek: DeepSeek V3deepseek83.052%
  7. OpenAI: GPT-4o (2024-11-20)openai82.835%
  8. DeepSeek: R1 0528deepseek82.809%
  9. Claude Sonnet 4anthropic82.47%
  10. Google: Gemini 2.5 Flash Litegoogle81.769%
  11. Qwen: Qwen3 Next 80B A3B Thinkingqwen80.737%
  12. Llama-4-Maverick-17B-128E-Instruct-FP8meta-llama79.969%

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