Benchmark profile

SWE-Atlas Refactoring

A Scale SWE-Atlas software-engineering agent benchmark focused on refactoring tasks.

How BenchLM shows SWE-Atlas Refactoring

BenchLM mirrors the public Scale SWE-Atlas Refactoring leaderboard from July 27, 2026 snapshot. The source reports 14 agent/model rows with confidence intervals and harness labels such as Claude Code, Codex, Gemini CLI, and Mini-SWE-Agent.

SWE-Atlas Refactoring is display only on BenchLM. It is useful evidence about software-engineering agents, but the rows mix base model quality with agent harness choices, so BenchLM keeps it out of weighted model-only rankings.

14 model variantsSWE-Atlas task familyRefactoring scoreScale Labs sourceDisplay only

SWE-Atlas Refactoring leaderboard SWE-Atlas paper

Refactoring score on SWE-Atlas Refactoring — July 27, 2026 snapshot

BenchLM mirrors the published refactoring score view for SWE-Atlas Refactoring. Fable-5 (Claude Code) xHigh leads the public snapshot at 54.8% , followed by Claude Opus 4.7 (Adaptive) (48.6%) and Opus 4.8 (Claude Code)\n (46.7%). BenchLM does not use these results to rank models overall.

Fable-5 (Claude Code) xHigh

Anthropic

54.8%

Overall —

2Closed

Claude Opus 4.7 (Adaptive)

Anthropic

Opus-4.7 (Claude Code)

48.6%

Overall 71.53Context 1M

Opus 4.8 (Claude Code)\n

Anthropic

46.7%

Overall —

14 modelsAgenticCurrentDisplay onlyUpdated July 27, 2026 snapshot

Refactoring score table (14 models)

Score

Fable-5 (Claude Code) xHighAnthropic

54.8%

Claude Opus 4.7 (Adaptive)Anthropic · Closed

48.6%

Opus 4.8 (Claude Code)\nAnthropic

46.7%

GPT-5.5OpenAI · Closed

44.8%

GPT-5.4OpenAI · Closed

44.3%

GLM-5.2Z.AI · Open weight

42.4%

GPT-5.3 CodexOpenAI · Closed

42.4%

Claude Opus 4.6Anthropic · Closed

35.6%

Gemini 3.1 ProGoogle · Closed

33.8%

Claude Sonnet 4.6Anthropic · Closed

32.2%

GLM-5Z.AI · Open weight

24.2%

Kimi K2.5Moonshot AI · Open weight

20.9%

MiniMax M2.5MiniMax · Closed

19.5%

Gemini 3 FlashGoogle · Closed

10.0%

The published SWE-Atlas Refactoring snapshot places Fable-5 (Claude Code) xHigh first at 54.8%. The third row is 8.1 points behind. The broader top-10 range is 22.5 points, so the table still separates the published systems.

14 models have been evaluated on SWE-Atlas Refactoring. The benchmark falls in the Agentic category. This category carries a 22% weight in BenchLM.ai's overall scoring system. SWE-Atlas Refactoring is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About SWE-Atlas Refactoring

Year

2026

Tasks

SWE-Atlas refactoring tasks

Format

Refactoring score with confidence intervals

Difficulty

Real-world software-engineering agent tasks

BenchLM mirrors the public Scale SWE-Atlas Refactoring leaderboard as a display-only agentic software-engineering benchmark. The source compares model-agent combinations such as Claude Code, Codex, Gemini CLI, and Mini-SWE-Agent.

SWE-Atlas Public benchmark source

BenchLM freshness & provenance

Version

SWE-Atlas Refactoring 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

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.

FAQ

What does SWE-Atlas Refactoring measure?

A Scale SWE-Atlas software-engineering agent benchmark focused on refactoring tasks.

Which model leads the published SWE-Atlas Refactoring snapshot?

Fable-5 (Claude Code) xHigh currently leads the published SWE-Atlas Refactoring snapshot with 54.8% refactoring score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on SWE-Atlas Refactoring?

14 AI models are included in BenchLM's mirrored SWE-Atlas Refactoring snapshot, based on the public leaderboard captured on July 27, 2026 snapshot.

Last updated: July 27, 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.