SWE-Marathon

SWE-Marathon - long-horizon software engineering benchmark. Evaluated by Abundant AI with 1M context length, max effort level, and 128K maximum output tokens.

Category

Agentic Coding

Max Score

100.0

Score Type

percent

Active

Yes

Model Scores

Scores for SWE-Marathon

Model Score Date Verified
GLM-5.2
24.49%
17.06.2026 Verified
Gemini 3.1 Pro
6.12%
17.06.2026 Unverified
GLM-5.1
1.00
17.06.2026 Unverified
GPT-5.5
22.45%
17.06.2026 Unverified
Claude Opus 4.8
51.02%
17.06.2026 Unverified
GLM-5.3
84.69%
28.08.2026 Verified
Kimi K3
96.12%
28.08.2026 Verified
Opus-4.8
97.55%
28.08.2026 Verified
Fable-5
65.51%
28.08.2026 Verified
GPT-5.6 Sol
84.69%
28.08.2026 Verified
Hy3
8.16%
28.08.2026 Verified
Hy4 preview
63.06%
28.08.2026 Verified
DeepSeek V4 Pro
36.73%
28.08.2026 Verified
Qwen3.8-2.4T-A95B
61.22%
28.08.2026 Verified
Claude Opus 5
100.00%
28.08.2026 Verified