Did GPT-5.2 actually outperform Claude Opus 4.5 in coding and reasoning benchmarks?

Yes—based on Fireship’s review, GPT-5.2 outperformed Claude Opus 4.5 on certain coding and reasoning benchmarks, specifically noting that GPT-5.2 “beats Claude Opus 4.5 on software engineering and reasoning.” The Fireship video also emphasizes a dramatic ARC/GPT-5.2 efficiency gain and frames coding accuracy as an area where the model shows fewer hallucinations in practice, though it cautions that day-to-day improvements may still be subtle for average users. By contrast, AI Explained’s coverage of Opus 4.5/4.6 focuses on broader benchmark variances and deployment realities, not a direct side‑by‑side scorecard that undercuts GPT-5.2’s performance claim in coding. Taken together, the sources suggest GPT-5.2 can outperform Opus 4.5 in specific coding-reasoning tasks, but the superiority is nuanced and bench‑to‑bench dependent. As always with AI benchmarks, context like task choice, time budget, and data windows heavily shapes the outcome. The consensus in the videos is to view these results as part of a shifting landscape rather than a single definitive win across all domains.

• Fireship points out that GPT-5.2 outperforms Claude Opus 4.5 on software engineering and reasoning tasks, highlighting a competitive edge in coding benchmarks.
• Fireship emphasizes a dramatic ARC benchmark improvement for GPT-5.2, framing the model as having fewer hallucinations in coding-related tasks, albeit with caveats about real-world day-to-day gains.
• AI Explained notes that benchmark results are sensitive to factors like thinking time and token budgets, implying that the apparent edge for GPT-5.2 can shift with resource constraints and task selection.
• AI Explained’s coverage of Opus 4.5/4.6 stresses that no model dominates across all benchmarks, so the GPT-5.2 advantage on coding/reasoning may not generalize to every metric or real-world scenario.