SWE-QA (Multi-Hop Code) - Interacting Entities: leaderboard

SWE-QA (2026) asks multiple-choice questions that need several pieces of code from real Python repositories, with the relevant code chunks supplied; questions and distractors were generated with a small LLM. Accuracy on questions about two entities interacting across three code chunks, in percent (chance 25); higher is better.

Metric: Accuracy (%) on the 4,488 Interacting-Entity questions, which follow two entities interacting across three chunks, of SWE-QA, multi-hop multiple-choice questions (four options) about the source of 12 SWE-bench Python repositories, generated with Llama-3.2-3B-Instruct; oracle setting (only the relevant code chunks), zero-shot, labels after the 66-item consensus correction; chance 25; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Llama 3.3 70B Instruct69.98#1385
2Gemma 3 4B (IT)68.6#4139
3Qwen 3 4B Instruct65.41#1696
4Qwen3-4B-Thinking (SWE-QA checkpoint unspecified)54.81
5GPT-OSS-20B53.83#1183
6SWE-QA DeepSeek-R1 (checkpoint unspecified)53.58
7Phi-4-mini (Thinking)50.35
8Phi-4 Mini Instruct47.9#6631
9Qwen 3 1.7B43.18#5860
10DeepSeek R1 Distill Qwen 1.5B39.14#9215
11SmolLM2-1.7B-Instruct29.94#9154
12SmolLM2-360M-Instruct20.61#9529

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

Interactive version: theaggregate.ai/benchmark?slug=swe-qa-multi-hop-code-interacting-entities · How It Works · Data refreshed daily, snapshot 2026-11-21.