R2ABench (Mini-SWE-Agent) - Edge F1: leaderboard

Requirements-to-architecture generation on R2ABench (2026): from a structured requirements specification of 68 educational and GitHub projects, the model driving the Mini-SWE-Agent framework writes a PlantUML architecture view, compared with a human-validated reference. F1 of the generated dependency and data-flow relations under semantic node alignment, over the views that parse, from 0 to 1; higher is better.

Metric: Edge F1 (0-1): F1 of generated dependency and data-flow relations against the reference view under semantic node alignment (L1), over the 68 R2ABench projects (17 educational and 51 GitHub repositories), generating a PlantUML architecture view from the structured requirements specification with the Mini-SWE-Agent framework under default settings and greedy decoding; scored only over the outputs that render and parse (the L0 gate), failed outputs excluded; higher is better. Source: arxiv.org. Saturation forecast: Around July 2028. 4 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-50.132#178
2Claude Sonnet 4.60.1243#173
3Qwen 3 Coder 480B A35B0.106#709
4DeepSeek V3.20.0704#419

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=r2abench-mini-swe-agent-edge-f1 · How It Works · Data refreshed daily, snapshot 2026-11-21.