FrontierCode 1.1 Extended

FrontierCode 1.1 Extended is a scorecard that lines up the performance and cost of AI systems that do coding work for you, side by side, so you can compare them. If a school exam only shows you the pass rate, this table shows the pass rate plus the exam fee it took to sit that test.

Inside the table, two numbers always come as a pair. One is a score showing how well a system solved medium-difficulty coding problems, and the other is the actual cost of solving one of those problems. So looking at this table lets you see at a glance which model or setup delivers similar skill for less money, or which ones cost about the same but differ in skill.

Because a coding agent calls the underlying model dozens or even hundreds of times to complete a single task, even a small drop in per-task cost has a big effect on overall running costs. So this benchmark isn't just about who's smarter — it's used to show who's smarter for the money.

The article states, "Looking at the FrontierCode 1.1 Extended benchmark table the company released, Fable 5 scored 62.8 on medium-difficulty tasks at $5.84 per task, while Fable 5.1 scored 63.6 at $2.68." The name alone might sound like a specific company's commercial product, but it actually refers to a benchmark results table that Cognition released to compare its own models and harnesses.