According to Greg Brockman on X, GPT-5.4 Pro (xhigh) achieved a 30% score on the CRITP research-level physics benchmark, up from a top score of 9% in November 2025, indicating a 10-point improvement and rapid gains in scientific reasoning (source: Greg Brockman on X). According to Haider (@slow_developer) cited in the same thread, progress is “way faster than expected,” underscoring improved multi-step derivations and symbol-heavy problem solving that are core to research workflows (source: Haider on X). As reported by the X thread, this trajectory aligns with OpenAI’s stated goal of building agents capable of conducting real research and discovering new scientific insights, signaling near-term opportunities for lab automation, theorem checking, and simulation-driven hypothesis generation in physics and adjacent domains (source: Greg Brockman on X).