logoalt Hacker News

SOLAR_FIELDS • today at 6:42 PM • 3 replies • view on HN

GPT models in my anecdotal experience are worse at drifting into the weeds on long running tasks, and continue to take what you ask way too literally. I have come to the conclusion that Anthropic’s models are just better at teasing out and intuiting what you actually mean even if you under specify it.

They are much better compute for dollar though. I have hilariously started to just have an opus thread maintain Sol subagents to get the best of both worlds


Replies

anakaine • today at 10:05 PM

Ive also had Opus coordinating multiple models, including those from Open AI, Deepseek, and Kimi. The cost of Opus is still my downfall, but the current setup is still far, far cheaper than doing everything in a frontier model, and more efficient than running cheaper models since the coordinator will take control when cheaper models repeatedly fail a task.

jpardoe • today at 6:52 PM

That’s exactly what I do. It’s more complex than this, but vaguely: Opus medium as orchestrator and reviewer, Sol(high/medium)/Terra(max) as implementer. My set up looks at my next GPT reset date and dynamically enters FAST mode if I’m under achieving on projected usage.

ctbellmar • today at 6:53 PM

Similar experience here. The spirals on long running tasks are out of control at times with hours going by without a conclusion.

➕ show 1 reply