Real-SWE
AssessTools
A benchmark for testing AI coding models on private, real-world enterprise repositories and tasks.
Why it's here
Placed in Assess: 2 article(s) of evidence from 2 source(s), led by research-stage coverage, with 2 in the last 30 days. Confidence 37%.
Evidence (2)
- 7The New Stack·9/14/2026researchReal-SWE benchmark shows top coding agents fail most of the time
The Real-SWE benchmark tested coding agents on private, real-world company codebases and found that Claude Fable 5.1 led with only 38.8% success, still failing more than 60% of attempts. Results varied sharply by model and task, and no agent was consistently reliable across unfamiliar production code.
- 7Hacker News·9/12/2026researchReal-SWE benchmarks AI agents on private enterprise codebases
Real-SWE is a new benchmark for evaluating frontier AI coding models on private, real-world enterprise codebases. It measures how well models handle proprietary code, company-specific rules, and production tasks with business impact, using native harnesses and pass@1-style resolution rates.