I gave Claude, ChatGPT, and a local 9B model the same 20 tasks, and only one was worth paying for

In 2027, both Anthropic’s Claude and OpenAI’s ChatGPT present a strong case for subscription services. The reporter personally found ChatGPT more useful due to its higher output within usage limits, but wonders if paying for Claude or ChatGPT is necessary. To investigate, the reporter gave Claude, ChatGPT, and a 9B local model named Qwen3.5-9B performing 20 tasks including coding, troubleshooting, and automation. The reporter compared accuracy, usability, and guidance needed to complete tasks.

For smaller tasks like explaining configuration settings, troubleshooting errors, and generating short functions, the local model performed reasonably well. However, when tasks required more reasoning or a longer conversation, the local model struggled. Claude and ChatGPT, on the other hand, excelled at these longer tasks.

In coding examples, Claude and ChatGPT produced usable code with fewer corrections needed. The local model also produced usable code but required multiple corrections.

When troubleshooting a Docker container, Claude provided clearer guidance for investigating the configuration and logs. ChatGPT also gave useful troubleshooting help, while the local model struggled with outdated instructions.

For app development tasks, Claude effectively handled connected changes and needed fewer corrections compared to ChatGPT. The local model required breaking the feature into smaller changes and taking responsibility for their integration.

Claude also demonstrated better results in using browsing tools for research tasks, while the local model needed more supervision to find and utilize information. Claude’s limited usage proved frustrating, but the usable results justified the subscription even with a free local model available.