[KongchangAI]
BenchmarkKingsbench 4

Kingsbench 4

A benchmark suite for evaluating the ability of AI models to build complete applications, comprising six full application-building tasks and belonging to the category of application-development evaluations.

Timeline (last 90 days)

Nov 19

In the test of 6 single-prompt items in Kingsbench 4, Carbon scored 32/60 overall (53.3%), ranking 8th among 13 models; Fable 5.2 ranked first with 70%; Carbon's base model Argon scored 54.3%.

Unverified50%
Nov 19

In the Kingsbench 4 hands-on test, Carbon's six builds consumed a total of 76 million tokens, about $203 at official list prices; Fable consumed 212 million tokens, about $222.

Unverified50%

All Facts (2)

Source Articles