SWE-AGI (Easy): leaderboard
Metric: Tasks passed (out of 6): the easy-tier SWE-AGI tasks whose final submission compiles and passes every hidden private test in a clean checkout; agents build the system in MoonBit from the task statement, normative specifications (RFCs, standards) and a fixed declaration-first API scaffold, iterating against public tests (10 percent) and submitting until the hidden private tests pass; each model runs in its own front-end (Codex CLI, Claude Code, Gemini CLI or Kimi CLI), with no budget limit; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 6 | #60 |
| 2 | Claude Opus 4.5 | 6 | #79 |
| 3 | GPT-5.3 Codex (xHigh) | 6 | #68 (GPT-5.3 Codex) |
| 4 | GPT-5.2 Codex (High) | 6 | #89 (GPT-5.2 Codex) |
| 5 | Gemini 3 Flash | 2 | #93 |
| 6 | GLM-4.7 | 2 | #185 |
| 7 | DeepSeek V3.2 (Thinking) | 1 | #198 (DeepSeek V3.2) |
| 8 | Claude Sonnet 4.5 | 0 | #138 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=swe-agi-easy · How It Works · Data refreshed daily, snapshot 2026-10-11.