MMLU Pro (Vals 5-Shot CoT)
General QA · 2024-06-03
Vals AI evaluates the full 14-subject MMLU-Pro benchmark using five in-context examples per subject and a chain-of-thought prompt. Its reported overall score is the unweighted mean of the 14 subject accuracies, rather than a micro-average over all 12,032 questions, so it is represented as an alternate component of the canonical MMLU-Pro factor family instead of independent evidence. Its guessing floor is the unweighted mean of the 14 subject-specific option-count expectations from official dataset revision b189ec765aa7ed75c8acfea42df31fdae71f97be.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5.1 | 92.4 |
| Gemini 3.1 Pro Preview | 91.0 |
| Gemini 3.8 Flash | 90.2 |
| Gemini 3.7 Flash | 90.1 |
| Gemini 3 Pro Preview | 90.1 |
| Opus 4.7 | 89.9 |
| Opus 4.8 | 89.6 |
| Gemini 3.5 Flash | 89.5 |
| Grok 4.6 | 89.4 |
| Qwen3.7-Max (source unspecified) | 89.3 |
| Gemini 3.6 Flash | 89.3 |
| Grok 4.5 | 89.2 |
| Opus 4.6 | 89.1 |
| GPT-5.6 Sol | 89.1 |
| Muse Spark 1.1 | 88.7 |
| Qwen3.8-Max | 88.6 |
| Gemini 3 Flash Preview | 88.6 |
| Muse Spark 1.2 | 88.3 |
| GPT-5.5 | 88.1 |
| Kimi K3 | 88.0 |
| Opus 4.1 | 87.9 |
| Qwen3.6 Plus (2026-04-02) | 87.7 |
| Kimi K2.6 | 87.6 |
| Sonnet 5 | 87.5 |
| GPT-5.4 | 87.5 |
| Sonnet 4.5 | 87.4 |
| Sonnet 4.6 | 87.3 |
| Muse Spark | 87.3 |
| Opus 4.5 | 87.3 |
| DeepSeek-V4-Pro | 87.2 |
| Qwen3.5 Plus (2026-02-15) | 87.2 |
| MiniMax M2.1 | 87.0 |
| DeepSeek V4 Pro 0813 | 87.0 |
| GLM-5.1 | 86.9 |
| GLM-5.3 | 86.8 |
| GLM-5.2 | 86.7 |
| GPT-5.6 Terra | 86.7 |
| GPT-5 | 86.5 |
| GPT-5.1 | 86.4 |
| Inkling | 86.3 |