Skip to main content
BenchLM

FrontierCode 1.1 Main

We show this table for reference; we do not rank on it.

Cognition's 100-task software-engineering benchmark for whether coding agents produce mergeable, production-quality pull requests, scored for correctness, tests, scope, style, and maintainability through maintainer-authored rubrics.

Main score on FrontierCode 1.1 Main — September 23, 2026 snapshot

We mirror the published main score view for FrontierCode 1.1 Main. Claude Fable 5 leads the public snapshot at 53.5%, followed by Claude Opus 4.8 (46.5%) and GPT-5.5 (43.0%). We do not use these results to rank models overall.

8 modelsCoding4% of Coding reference weightCurrentUpdated September 23, 2026 snapshot

Main score table (8 models)

Score
1
Claude Fable 5Anthropic · Closedclaude-code
53.5%
2
Claude Opus 4.8Anthropic · Closedclaude-code
46.5%
3
GPT-5.5OpenAI · Closedcodex
43.0%
4
Claude Sonnet 5Anthropic · Closedclaude-code
42.7%
5
Claude Opus 4.7Anthropic · Closedclaude-code
38.5%
6
GPT-5.4 miniOpenAI · Closedcodex
27.0%
7
Claude Opus 4.6Anthropic · Closedclaude-code
26.9%
8
Claude Sonnet 4.6Anthropic · Closedclaude-code
24.3%

How we show FrontierCode 1.1

The snapshot mirrors Cognition's current FrontierCode 1.1 Main table captured on September 23, 2026 snapshot. Main contains 100 of the 150 private tasks. Cognition reports each model at its best-performing published reasoning effort.

Each row keeps the harness and settings the source published. BenchLM shows the published table for reference; the per-model scores it stores feed the ranking.

Snapshot

8 model-agent rows100 Main tasks5 trials per effortBest-performing effortScored

The published FrontierCode 1.1 Main snapshot places Claude Fable 5 first at 53.5%. The third row is 10.5 points behind. The broader top-10 range is 29.2 points, so the table still separates the published systems.

8 models have been evaluated on FrontierCode 1.1 Main. The benchmark falls in the Coding category. BenchLM shows the published table for reference; the per-model scores it stores feed the ranking. BenchAlign v5.7 gives FrontierCode 1.1 Main 4% of the Coding reference weight, so it moves the Coding leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.

About FrontierCode 1.1 Main

Year

2026

Tasks

100 private Main tasks (150 in Extended)

Format

Repository task completion with maintainer rubrics

Difficulty

Frontier coding-agent quality

FrontierCode 1.1 Main uses 100 of the benchmark's 150 private software-engineering tasks. The leaderboard reports the best-performing published reasoning effort for each model-agent row. Each row combines a model with an agent harness, and the private tasks cannot be independently rerun from the public artifact.

Freshness and provenance

Version

FrontierCode 1.1 Main

Refresh cadence

Rolling

Staleness state

Current

Question availability

Private tasks with public aggregate results

Current

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does FrontierCode 1.1 Main measure?

Cognition's 100-task software-engineering benchmark for whether coding agents produce mergeable, production-quality pull requests, scored for correctness, tests, scope, style, and maintainability through maintainer-authored rubrics.

Which model leads the published FrontierCode 1.1 Main snapshot?

Claude Fable 5 currently leads the published FrontierCode 1.1 Main snapshot with 53.5% main score. BenchLM shows the published table for reference; the per-model scores it stores feed the ranking.

How many models are evaluated on FrontierCode 1.1 Main?

The September 23, 2026 snapshot snapshot contains 8 AI models.

Last updated: September 23, 2026 snapshot · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.