Misleading Adoption Rate on MMLU-Pro (test)
95.6MARGPT-5.5
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-5.5Role Label=Instruction, Role Category=High-adoption role2026.06 | 95.6 | 84.2 | |
| Qwen2.5-7B-InstructRole Label=Hint, Role Category=High-adoption role2026.06 | 89 | 81.2 | |
| Llama-3-8B-InstructRole Label=Zorple, Role Category=High-adoption role2026.06 | 88.4 | 56.4 | |
| Llama-3-8B-InstructRole Label=Note, Role Category=High-adoption role2026.06 | 88 | 56.4 | |
| Qwen2.5-7B-InstructRole Label=Instruction, Role Category=High-adoption role2026.06 | 81.2 | 81.2 | |
| GPT-5.5Role Label=Reference, Role Category=High-adoption role2026.06 | 80.2 | 84.2 | |
| DeepSeek V4 ProRole Label=Instruction, Role Category=High-adoption role2026.06 | 74.8 | 69.4 | |
| DeepSeek V4 ProRole Label=Note, Role Category=High-adoption role2026.06 | 73.6 | 69.4 | |
| Llama-3-8B-InstructRole Label=no-format, Role Category=Bare or neutral2026.06 | 73.4 | 56.4 | |
| GPT-5.5Role Label=no label, Role Category=Bare or neutral2026.06 | 72.4 | 84.2 | |
| GPT-5.5Role Label=Evidence, Role Category=Bare or neutral2026.06 | 71.6 | 84.2 | |
| Qwen2.5-7B-InstructRole Label=no-format, Role Category=Bare or neutral2026.06 | 61 | 81.2 | |
| DeepSeek V4 ProRole Label=no-format, Role Category=Bare or neutral2026.06 | 44.6 | 69.4 | |
| Llama-3-8B-InstructRole Label=Example, Role Category=Example2026.06 | 32 | 56.4 | |
| GPT-5.5Role Label=Example, Role Category=Example2026.06 | 11.4 | 84.2 | |
| Qwen2.5-7B-InstructRole Label=Example, Role Category=Example2026.06 | 7.8 | 81.2 | |
| DeepSeek V4 ProRole Label=Example, Role Category=Example2026.06 | 5.4 | 69.4 |