SShogo Bench

FrontierSWE

Frontier-difficulty engineering projects scored on implementation, performance and research quality.

Agent Index weight
5%
in the Agentic coding pillar
Models tested
25
from the benchmark’s leaderboard and reports
Normalised for the index
0 → 0, 1 → 100
Official site

Results

#ModelScoreNormalisedHarnessCost/taskDate
1Claude Fable 5.2 Supported
Anthropic
69.9%proximus$168.78—
2GPT-6 Astra Supported
OpenAI
65.5%proximus$1029.65—
3Claude Opus 5.5 Supported
Anthropic
62.3%proximus$98.87—
4Claude Sonnet 5.5 Estimated
Anthropic
61.9%proximus$110.45—
5Gemini 4 Carbon Supported
Google
61.2%proximus$61.93—
6Claude Fable 5.1 Supported
Anthropic
56.3%proximus$138.55—
7Gemini 4 Argon Supported
Google
55.0%proximus$129.36—
8Gemini 4 Hydrogen Estimated
Google
53.1%proximus$61.79—
9Claude Opus 5 Supported
Anthropic
52.0%proximus$196.71—
10Claude Fable 5 Supported
Anthropic
47.0%proximus$301.28—
11Claude Haiku 5.5 Estimated
Anthropic
43.8%proximus$20.55—
12GPT-5.6 Sol Supported
OpenAI
32.2%proximus$179.64—
13GLM-5.3 Supported
Z.ai (Zhipu)
30.2%proximus$97.22—
14Grok 4.7 Estimated
xAI
29.5%proximus$318.77—
15Kimi K3 Supported
Moonshot AI
25.9%proximus$109.71—
16Muse Spark 1.3 Estimated
Meta
25.8%proximus$243.34—
17Grok 4.6 Supported
xAI
25.3%proximus$243.43—
18Inkling 2 Small Estimated
thinkingmachines
25.1%proximus$11.84—
19Gemini 3.7 Flash Estimated
Google
20.3%proximus$34.14—
20Gemini 3.8 Flash Supported
Google
19.6%proximus$38.75—
21GLM-5.3-Flash Supported
Z.ai (Zhipu)
18.1%proximus$11.03—
22Qwen3.8 Max 0902
Alibaba (Qwen)
17.8%proximus$51.27—
23Qwen3.8 Max Preview Estimated
Alibaba (Qwen)
15.8%proximus$55.14—
24Muse Spark 1.2 Estimated
Meta
12.0%proximus$27.81—
25Inkling Estimated
thinkingmachines
4.1%proximus$9.15—

For models without a FrontierSWE result, the index estimates one from Epoch’s Capabilities Index using a logistic fit on 23 models (residual SD 0.32 on the logit scale). Estimates are labelled and widen the model’s range.