Semantic Uncertainty Stability Prediction on BBEH
97DCS AUROCOutput Embedding
Evaluation Results
| Method | Links | |
|---|---|---|
| Output EmbeddingLLM=Sky-OR1 32B2026.05 | 97 | |
| Output EmbeddingLLM=Llama 8B2026.05 | 96 | |
| Output EmbeddingLLM=DS-R1 32B2026.05 | 96 | |
| Last Input TokenLLM=Sky-OR1 32B2026.05 | 96 | |
| Last Input TokenLLM=DS-R1 32B2026.05 | 95 | |
| Soft ThinkingLLM=DS-R1 32B2026.05 | 95 | |
| Soft ThinkingLLM=Sky-OR1 32B2026.05 | 95 | |
| Last Input TokenLLM=Llama 8B2026.05 | 94 | |
| Soft ThinkingLLM=Llama 8B2026.05 | 94 | |
| Input EmbeddingLLM=DS-R1 32B2026.05 | 94 | |
| Input EmbeddingLLM=Llama 8B2026.05 | 93 | |
| Input EmbeddingLLM=Sky-OR1 32B2026.05 | 93 | |
| Latent ThinkingLLM=Llama 8B2026.05 | 92 | |
| Input EmbeddingLLM=Llama 70B2026.05 | 92 | |
| Latent ThinkingLLM=DS-R1 32B2026.05 | 92 | |
| Latent ThinkingLLM=Sky-OR1 32B2026.05 | 92 | |
| Soft Thinking with NoiseLLM=Llama 8B2026.05 | 90 | |
| Output EmbeddingLLM=Llama 70B2026.05 | 89 | |
| Last Input TokenLLM=Llama 70B2026.05 | 89 | |
| Latent ThinkingLLM=Llama 70B2026.05 | 87 | |
| Soft Thinking with NoiseLLM=Sky-OR1 32B2026.05 | 86 | |
| Soft ThinkingLLM=Llama 70B2026.05 | 85 | |
| Soft Thinking with NoiseLLM=DS-R1 32B2026.05 | 85 | |
| Soft Thinking with NoiseLLM=Llama 70B2026.05 | 84 | |
| Output EmbeddingLLM=GPT-OSS 20B2026.05 | 74 | |
| Latent ThinkingLLM=GPT-OSS 20B2026.05 | 59 | |
| Input EmbeddingLLM=GPT-OSS 20B2026.05 | 59 | |
| Last Input TokenLLM=GPT-OSS 20B2026.05 | 58 | |
| Random VectorLLM=GPT-OSS 20B2026.05 | 56 | |
| Soft ThinkingLLM=GPT-OSS 20B2026.05 | 55 | |
| Random VectorLLM=Llama 8B2026.05 | 52 | |
| Random VectorLLM=Llama 70B2026.05 | 50 | |
| Random VectorLLM=DS-R1 32B2026.05 | 50 | |
| Random VectorLLM=Sky-OR1 32B2026.05 | 49 | |
| Soft Thinking with NoiseLLM=GPT-OSS 20B2026.05 | 46 |