It claimed the "best at coding" crown
in Tokyo.
Anthropic held its "Code with Claude" developer conference in Tokyo for the first time. In the keynote it positioned Fable 5 as "the highest-performing model for coding," raising a benchmark 80% success rate as its banner. Tokyo is the third city after San Francisco and London—we unpack what that means in diagrams.
It "said it outright" in Tokyo—
about an already-announced model
The Fable 5 model itself had already been announced. What made news this time was where and how its performance was repositioned. Anthropic held its "Code with Claude" developer conference in Tokyo for the first time, and in that keynote it officially declared Fable 5 "the highest-performing model for coding."
The symbol of it all was the figure of an 80% success rate on a coding benchmark. What matters is that strengths described piecemeal until now were consolidated, in Tokyo, into a single banner: "the 80% coding model." This is less an update to the performance than an update to who it is being addressed to.
| Until now | This time (Tokyo) |
|---|---|
| A developer approach centered on SF and London | Direct to Asia-Pacific and the Japanese-speaking world |
| Strengths described piecemeal | Consolidated into a single number, "80%" |
| Almost no official events for the Japanese-speaking world | First Tokyo event (third city) |
| Fable 5 already announced | Officially positioned as "the highest-performing" |
Not an announcement of performance.
It redrew "which developers to reach."
The three strengths
behind the "80%"
In the keynote, the strengths of a coding model were organized into three. The emphasis is not on one-off brilliance but on the ability to keep running correctly over the long haul.
The first is prediction accuracy on a single prompt—the ability to get close to the intended code in one shot, even from a short instruction. The second is consistency on large-scale, long-term projects (the time horizon)—keeping the context in view and not letting the direction drift even on work that runs on and on.
The third is code comprehension. Not just writing, but understanding existing code, pinpointing the cause of a bug, and going as far as proposing improvements—in other words, the "read and fix" side of the ability. Bundle the three together and it becomes clear that what Fable 5 aims for is not quick reflexes but a character that carries a big, long project all the way to the end.
Why was it "Tokyo"?
The choice of location is itself a message. Anthropic is beginning to reach out to developer communities outside the US—the Japanese-speaking world in particular.
Until now, centered on SF and London
Anthropic's developer-facing approach has centered on the English-speaking world, with almost no official events aimed at Japanese-speaking developers. In reality, information skewed toward those who could follow English-language materials.
Choosing Tokyo as the third city
Placing Tokyo as the third venue after San Francisco and London was, in itself, a clear signal of valuing the Asia-Pacific and Japanese markets.
Direct outreach to the Japanese-speaking world
It is a large-scale, direct approach, and it also raises hopes that documentation and support will increasingly be localized into Japanese. Japanese engineers gained a sense that "this is being said to us."
Whether it matters to you, or not
There is a clear split between people for whom this announcement is a decision factor and people for whom it is not.
People working in Japanese
For those who were considering Claude for development involving Japanese specs and comments, this is a tailwind signal that "Anthropic values the Japanese market."
People undecided on tooling
For those torn between "Cursor or Claude Code," Anthropic's commitment to Japan adds one more factor to the decision.
People who have already decided
For those committed to a particular tool, or who have no trouble with English materials, nothing much changes for now. This is not a reason to rush a switch.
Read the stance, not the number
The headline "the 80% coding model" is striking, but the real news is not the number itself. It is that a frontier lab stepped squarely into developer communities outside the US and beyond the English-speaking world—and for Japanese engineers, that is the turning point.
Performance is not decided by a single benchmark, and the best tool differs from one workplace to the next. That is precisely why, beyond "which model is faster," the axis of how seriously the provider regards your language and your market will gradually matter in future selection. The declaration in Tokyo nudged that axis up a notch.