https://korchasa.dev/tags/benchmarks/Recent content in Benchmarks on korchasa@*opsHugoenFri, 12 Jun 2026 00:00:00 +0000https://korchasa.dev/posts/2026_06_12_agents_comparison_benchmarks/Fri, 12 Jun 2026 00:00:00 +0000https://korchasa.dev/posts/2026_06_12_agents_comparison_benchmarks/<p><a href="https://github.com/korchasa/flowai-experiments/tree/main/agents-comparison" rel="nofollow noopener noreferrer external">https://github.com/korchasa/flowai-experiments/tree/main/agents-comparison</a></p> <p><img src="https://korchasa.dev/posts/2026_06_12_agents_comparison_benchmarks/image.png" alt="Illustration: autonomous agent benchmarks"></p> <p>Burned 40% of my limit running benchmarks close to my real tasks on opus/fable/gpt-5.5 — fully autonomous agent work: app generation from scratch, a project audit, and three implementation tasks of varying difficulty.</p> <p>What can be said with at least some confidence:</p> <ul> <li>fable beats opus-4.8 and gpt-5.5 on result quality. My working hypothesis: fable medium = opus xhigh.</li> <li>opus xhigh is unexpectedly expensive due to overly long reasoning. Sometimes more expensive than fable.</li> <li>Looks are still a pain. Everything is dark-neon-identical.</li> <li>Proper testing will take 1-2 weekly limits on claude x20.</li> </ul> <p>Hypotheses:</p>