CODE IN TOKYO

Anthropic and OpenAI Are Building Toward Different Visions of AI

This Is Not Simply a Question of Which Model Is Smarter

In my previous blog post, I wrote about my experience with GPT-5.6 Sol.

I said that Fable still seemed one step smarter. When I reread the article later, however, I realized that the wording was not quite accurate.

This is not simply a comparison of intelligence.

Anthropic and OpenAI appear to have different philosophies about the kind of AI they are trying to build.

Using both companies’ models every day has gradually made that distinction more visible to me.

Anthropic Appears to Be Pursuing a More Capable Intelligence

My impression is that Anthropic is trying to build an AI that is, in itself, more intelligent.

When I give Claude a complicated problem, it sometimes organizes the issue or proposes a design in a way that a human might not have considered.

I once gave Fable a particularly complex modification to an existing system. It spent more than an hour investigating the requirements, considering the implementation strategy, and working through the change. The result was good enough to adopt with almost no modification.

Honestly, it was the kind of work that would not have been easy for a human either.

That level of intelligence can also create problems. At times, Claude seems to go beyond the user’s intent and pursue what it considers the optimal solution.

When that instinct points in the right direction, the result can be remarkable.

When it does not, a human has to step in and say, “That is more than I asked for,” and guide it back on course.

OpenAI Appears to Be Building an AI That Works with People

OpenAI feels somewhat different to me.

It appears to be aiming for an AI that works alongside people.

Earlier OpenAI models required extremely detailed instructions. If I did not explain my assumptions and reasoning clearly enough, the output often exposed every point where I had hoped the model would infer what I meant.

For a long time, I therefore used OpenAI models mainly for supporting tasks and relied on Claude for architecture and difficult decisions.

Using Sol changed that impression.

Its basic character has not changed. It is still the kind of model that steadily carries out what it has been asked to do.

The difference is that it has become much better at understanding the intent behind instructions that are slightly ambiguous.

That improvement is more significant than I expected.

For everyday work such as implementation, investigation, and writing, I increasingly find myself wondering whether Sol is already sufficient.

Different Strengths for Different Problems

Even so, if I needed an AI to solve an unfamiliar problem, I would still choose Fable.

It continues to feel a step ahead in creativity and system design.

Rather than saying that Fable is simply smarter, however, I think it is more accurate to say that the two companies have been pursuing different goals.

Anthropic has focused on making the AI itself more intelligent.

OpenAI has developed an AI that collaborates with people and reliably finishes the work.

Looking at Sol, I also get the sense that these two directions are gradually moving closer together.

That is why I expect there will be fewer situations in which not having access to Claude becomes a serious problem.

The two competitors are beginning to absorb each other’s strengths.

As long as AI companies continue competing this way, users stand to benefit the most.

Contact

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

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