CODE IN TOKYO

I Tried GPT-5.6 Sol—and I Think It Is a Viable Alternative to Claude

Automating My Blog-to-YouTube Workflow with Sol

I recently tried OpenAI’s new model, GPT-5.6 Sol.

For this test, I asked it to build an end-to-end system that turns a blog article into a video and uploads it directly to YouTube.

My usual blogging workflow starts with me writing the original draft. Codex then turns it into an article ready for publication. This time, I extended that workflow by automating the process of generating a video from the finished article and uploading it to YouTube.

Overall, Sol handled the task without any major problems.

It understood what I wanted and implemented the system without going badly off track. For ordinary development work, I do not think it is meaningfully behind Claude Opus. It is already capable enough to serve as a practical alternative to Claude.

Fable Still Appears to Be a Step Ahead

That said, I currently feel that Claude Fable still stands above the others.

Recently, I gave Fable a particularly complicated modification to an existing feature. I asked it to investigate the current behavior, identify the affected areas, develop an implementation plan, and then carry out the change.

At one point, Fable spent about an hour and a half thinking. It took so long that I started wondering whether it had stalled, and I checked in occasionally to see what it was doing.

The final implementation, however, was highly satisfactory. It accounted for the complicated existing requirements and produced a coherent modification almost entirely in one pass.

If I had given the same task to Opus, I suspect the first implementation would have contained some questionable decisions. A human would then need to point out the problems, ask it to investigate again, and repeat the correction process several times. The entire task could easily take half a day.

I suspect Sol would follow a similar pattern.

It would probably complete the implementation eventually. But when it comes to getting close to the right answer on the first attempt, Fable still seems stronger.

The Difference Is Not Simply Code Generation

The distinction is not just about the ability to generate code.

The gap becomes visible when a task involves complicated requirements, a broad review of an existing codebase, ambiguous details that need to be resolved, and decisions about exactly what should be changed.

Fable is particularly strong at handling that entire process, including the investigation and design work that comes before implementation.

At the same time, not every task requires Fable.

For an automation system like the one I built this time, or for development work where the overall direction is relatively clear, Sol was more than sufficient. It looks like a very strong option for everyday work.

My Current Ranking

My current impression is that Fable sits at the top, with Sol and Opus in the tier below it.

Sol is not quite at Fable’s level. Even so, I expect there will be far fewer situations in which not having access to Claude becomes a serious problem.

At the very least, Sol is already a credible alternative to Claude.

I have also turned this article into a video and uploaded it. The synthetic narration is not ideal, but for now, I am not motivated enough to record the narration myself.

Contact

I provide system development and technical consulting using AI, AWS, and Claude Code.

Contact form: https://holly-money-e94.notion.site/390ff30cf18c8086a676fe630d171873