SWE-Arena: leaderboard

Software Engineering Arena: head-to-head Elo-rated coding evaluation where 37 models solve real software engineering tasks, judged by pairwise human preference.

Metric: Elo Score. Source: huggingface.co. Source: swe-arena.com. Source: huggingface.co. 37 models tracked.

Top models

#ModelScoreOverall rank
1Voxtral-Small-24B-25071004#581
2Claude 3.5 Sonnet1002#350
3DeepSeek V3.11002#264
4O31002#124
5GLM 4.5 Air1002#392
6Qwen 3 VL 30B A3B (Thinking)1002#351 (Qwen 3 VL 30B A3B)
7Grok 3 Mini1002#281
8ministral-3B-25121002#917
9Mistral 7B Instruct1002#1363
10Hermes-2-Pro-Llama-3-8B1002#1127
11GLM-4 32B1002#459
12ERNIE 4.5 VL 28B A3B1002#695
13GPT-4o1000#343
14Gemini 2.5 Pro (Preview 06-05)1000#106
15Gemini 2.5 Pro998#152

Interactive version: theaggregate.ai/benchmark?slug=swe-arena · How It Works · Data refreshed daily, snapshot 2026-10-20.