Inference
EN
Epoch AI

SWE-Bench verified

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.32
Top resultBest result on this benchmark.83.5%Claude Opus 4.7
Top-3 spreadPoints from first to third.4.1 pts
YearYear of release.—

Result and price

29 of 32 results

Results

32 results
#ModelResultThe score from the source.
01Claude Opus 4.7max effort83.5% ±1.7
02GPT-5.5xhigh effort80.6% ±1.8
03Gemini 3.5 Flashhigh effort79.3% ±1.8
04GLM-5.2max effort78.7% ±1.9
05DeepSeek V4 Pro 0423max effort77.6% ±1.9
06Qwen3.7 Max77.3% ±1.9
07Claude Opus 4.677.2% ±2.0
08GPT-5.4high effort76.9% ±1.9
09Kimi K2.676.7% ±1.9
09Qwen 3.6 Max (preview)76.7% ±1.9
09Claude Opus 4.576.7% ±1.9
12Gemini 3.1 Pro75.6% ±2.0
13Gemini 3 Flash75.4% ±2.0
14Claude Sonnet 4.675.2% ±2.0
15GPT-5.3 Codexhigh effort74.8% ±2.0
16GLM-5.174.2% ±2.0
17GPT-5.2high effort73.8% ±2.0
17Kimi K2.573.8% ±2.0
19GPT-5high effort73.6% ±2.0
20Claude Opus 4.173.3% ±2.0
21Gemini 3 Pro72.9% ±2.0
22GLM-572.1% ±2.1
23Claude Sonnet 4.571.3% ±2.1
24Claude Opus 470.7% ±2.1
25GPT-5.1high effort66.9% ±2.1
26GPT-5 minimedium effort64.7% ±2.2
27o3medium effort62.3% ±2.2
28Claude 3.7 Sonnet61.0% ±2.2
29Qwen3.6 Plus57.9% ±2.2
30Gemini 2.5 Pro (Jun 2025)57.6% ±2.3
31GPT-4.148.5% ±2.3
32GPT-4o31.0% ±2.1