Claude, ChatGPT and a local 9B model tested on 20 tasks: Which is worth paying for

As artificial intelligence tools evolve, deciding whether to pay for a subscription becomes crucial. A developer tested Anthropic's Claude, OpenAI's ChatGPT, and a local Qwen3.5-9B model running via LM Studio across 20 regular technical tasks, ranging from coding to automation.

• The evaluation covered 20 daily tasks including coding, Docker debugging, n8n workflows, and documentation research.
• The local 9B model handles narrow questions well but requires heavy supervision.
• ChatGPT offers great daily limits, but Claude excels in complex, multi-step engineering projects.

Coding and handling straightforward requests

For basic coding tasks, configuration explanations, and generating short functions, all three models performed adequately. The local model can produce usable code when provided with strict constraints and expected inputs, though human oversight remains essential for catching errors.

As tasks scale in complexity and require architectural decisions, the differences widen. ChatGPT and Claude handle requests with significantly less preparation from the user, whereas the local model requires considerable hand-holding to achieve similar outcomes.

Complex projects, tools, and web browsing

When tackling complex workflows—such as adding features to an application or integrating MCP server tools—Claude consistently outperforms the others. It effectively tracks related files, debugs logs, and uses browsing tools to analyze up-to-date documentation.

In contrast, the local 9B model struggles heavily with these extended sequences. The amount of supervision required to keep the local model on track often outweighs the convenience of handing it the task in the first place.

Conclusion and subscription value

Despite frustrating usage limits, Claude's superior ability to translate prompts directly into working solutions justifies its subscription cost for demanding technical projects.

While ChatGPT remains a versatile daily driver and local models offer absolute privacy and zero cost, Claude stands out as the only subscription truly worth paying for when tackling advanced, multi-layered development work.