Verdict
GPT-4 (Mar 2023) leads 3–1 across 4 shared benchmarks.
GPT-4 (Jun 2023) 1 · GPT-4 (Mar 2023) 3 · higher isn't always better — check the caveats below.
Reported scores — protocols may differ. How we compare
Math: Evenly split 1–1 across 2 shared math benchmarks (largest gap: GSM8K, 92 vs 89.99 — single evaluator).
GPT-4 (Mar 2023) leads 3–1 across 4 shared benchmarks. GPT-4 (Jun 2023) leads 1 benchmark and GPT-4 (Mar 2023) leads 3. Higher isn't always better — see the integrity caveats on each benchmark.
GPT-4 (Jun 2023) and GPT-4 (Mar 2023) have 4 benchmarks in common in our data — those are the only rows where a direct, apples-to-apples comparison is drawn.