Verdict
GPT-6 Astra leads 10–8 across 20 shared benchmarks.
Claude Opus 5.5 8 · GPT-6 Astra 10 · 2 tied · higher isn't always better — check the caveats below.
Reported scores — protocols may differ. How we compare
Unknown: Claude Opus 5.5 leads 4–3 across 7 shared unknown benchmarks (largest gap: SpireClimb, 25.67 vs 3.33 — protocols undisclosed).
Coding: Claude Opus 5.5 leads 2–1 across 3 shared coding benchmarks (largest gap: MirrorCode, 77.4 vs 46.7 — protocols undisclosed).
Reasoning: GPT-6 Astra leads 2–0 across 3 shared reasoning benchmarks (largest gap: GPQA diamond, 94.36 vs 87.46 — protocols undisclosed).
Agentic: Claude Opus 5.5 leads 2–0 across 2 shared agentic benchmarks (largest gap: APEX-Agents, 73.5 vs 64.7 — protocols undisclosed).
Math: GPT-6 Astra leads 1–0 across 2 shared math benchmarks (largest gap: FrontierMath-Tier-4-v2-Private, 97.6 vs 95 — protocols undisclosed).
GPT-6 Astra leads 10–8 across 20 shared benchmarks. Claude Opus 5.5 leads 8 benchmarks and GPT-6 Astra leads 10, with 2 tied. Higher isn't always better — see the integrity caveats on each benchmark.
Claude Opus 5.5 and GPT-6 Astra have 20 benchmarks in common in our data — those are the only rows where a direct, apples-to-apples comparison is drawn.
GPT-6 Astra leads reasoning 2–0. Claude Opus 5.5 leads agentic 2–0.