AI Evaluation
Cognition Opens FrontierCode Results to Public Inspection
Cognition made its FrontierCode leaderboard publicly browsable on July 18. The page evaluates coding agents on whether a maintainer would merge their patch, exposes sample tasks and zeros runs detected consulting solution-bearing sources.
What happened and why it matters
Public task inspection makes the benchmark more legible, while the scores remain Cognition's own evaluation rather than neutral industry consensus.
Primary source
Primary reference: Cognition: FrontierCode Leaderboard. Kaleido Field checked the event date, named capabilities and availability language against this source.
| Source date | Public leaderboard reported July 18, 2026; methodology revision July 7, 2026 |
|---|---|
| Checked by Kaleido Field | July 19, 2026, 09:18 CST |
| What this source supports | benchmark-maker leaderboard with public methodology and task inspection for Cognition FrontierCode public leaderboard mergeability rollout cost unfair internet use |
| What it does not prove | It does not prove a universal product ranking, full regional availability, or performance on every visual intelligence task. |
What the page exposes
The public page presents FrontierCode 1.1 results alongside methodology and an interactive task example. Cognition grades correctness, test quality, scope discipline, style and repository conventions, then asks whether the resulting pull request is mergeable.
Tasks are written with open-source maintainers and graded through tests, rubrics and other verifiers. The page says runs that consult original pull requests or other solution-bearing sources are detected and scored zero.
What changed from a static score
Readers can inspect a task brief, model patch and failed criteria rather than accepting one aggregate number. Cost per rollout also sits beside score, which helps separate expensive brute-force attempts from more efficient runs.
That transparency does not remove benchmark design choices. Maintainer selection, task mix, rubric interpretation, harness configuration and model access can all affect the ordering.
Evidence boundary
Cognition created FrontierCode and operates the leaderboard. Its July 7 revision changed internet-use handling and deprecated the Diamond subset, so comparisons should name the 1.1 revision rather than blending it with the original June release.
Kaleido Field treats the page as an official benchmark-maker listing, not an independent evaluation. A high FrontierCode score does not establish production reliability across every repository, language or organisation.
Evidence boundary
This page reports a dated event from a named primary source. Company specifications and adoption statements remain attributed claims unless independent evidence is cited above.
FAQ
What is the practical answer?
Cognition made its FrontierCode leaderboard publicly browsable on July 18. The page evaluates coding agents on whether a maintainer would merge their patch, exposes sample tasks and zeros runs detected consulting solution-bearing sources.
What source does this article use?
The primary source is Cognition: FrontierCode Leaderboard. Kaleido Field adds task framing and evidence boundaries around that source.
Where should the user verify the answer?
Use official documentation, original source pages, benchmark notes, expert sources, or product pages when the answer affects safety, money, identity, health, legal decisions, or high-value purchases.