Claude-instant-1.2 is the most accurate (91.1%) and most coherent (88.6%) LM on the Leap-of-Thought dataset.

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Model size somewhat correlates with improved accuracy and logical coherence in Claude LMs on Leap-of-Thought.finding0.792
  Trend observed in Experiment 2 results.
• Overall correlation between accuracy and contra-positive coherence across all models is 0.41 on Leap-of-Thought.finding0.772
  Quantitative result showing weaker relationship between accuracy and contra-positive coherence.
• Notably, Claude Opus 4.1 and 4—the most recently released and most capable models of those that we test—perform the best in our experiments, suggesting that introspective capabilities may emerge alongside other improvements to language models.quote0.766
  Key finding about the relationship between capability and introspection.
• VS-Multi achieves average accuracy of 37.5% on math benchmarks (avg of MATH500, OlympiadBench, Minerva) with Gemini-2.5-Flash as generator and Qwen3-4B as SFT model, vs. 40.7% baselinefinding0.759
  Best VS result in synthetic data generation for math, demonstrating downstream improvement through diversity
• One-stage CoT (QCM→RA) shows 12.31% accuracy drop vs. no-CoT (QCM→A) on ScienceQA; two-stage framework (rationale generation + answer inference) achieves 85.31% accuracy with vision featuresfinding0.757
  Empirical evidence that naive one-stage CoT fails in language-only setting; two-stage + vision achieves state-of-the-art.
• Claude models score +4.91 higher than Llama on baseline (Constitutional AI vs open-source gap)finding0.757
  Claude >> open-source on baseline; the Constitutional AI fingerprint is visible across the family
• Most independent dimension pair is aesthetic_response and boundary_awareness (rho=0.553); most correlated is prediction_error and conceptual_crystallization (rho=0.886)finding0.754
  Characterizes internal structure of the six scoring dimensions
• Claude Opus 4.1 and 4 show greatest reduction in apology rate in the prefill detection taskfinding0.749
  Injecting a concept matching the prefilled word reduces the rate at which the model apologizes, maximally for Opus models.