SShogo Bench

APEX-Agents

Long tasks from investment banking, consulting and corporate law, done with real office tools.

Agent Index weight
8%
in the Knowledge-work agents pillar
Models tested
62
from the benchmark’s leaderboard and reports
Normalised for the index
0 → 0, 1 → 100
Official site

Results

#ModelScoreNormalisedHarnessCost/taskDate
1Gemini 4 Carbon Supported
Google
84.2%source——
2Gemini 4 Argon Supported
Google
82.4%source——
3Claude Fable 5.2 Supported
Anthropic
80.1%source——
4Gemini 4 Hydrogen Estimated
Google
78.0%source——
5Claude Sonnet 5.5 Estimated
Anthropic
75.5%source——
6Claude Opus 5.5 Supported
Anthropic
73.5%source——
7Grok 4.7 Estimated
xAI
70.0%source——
8DeepSeek V4.1 Flash 1017 Estimated
DeepSeek
69.8%source——
9Claude Fable 5.1 Supported
Anthropic
68.6%source——
10GPT-6.1 Terra Estimated
OpenAI
68.3%source——
11Gemini 3.7 Flash Estimated
Google
67.8%source——
12Muse Wonder Estimated
Meta
66.1%source——
13Claude Opus 5 Supported
Anthropic
65.8%source——
14Kimi K3.1 Pro
Moonshot AI
65.5%source——
15Grok 4.6 Supported
xAI
65.3%source——
16GPT-6 Astra Supported
OpenAI
64.7%source——
17Gemini 3.8 Flash Supported
Google
64.3%source——
18Muse Ember
Meta
64.3%source——
19Claude Fable 5 Supported
Anthropic
63.6%source——
20Qwen3.8 Max Preview Estimated
Alibaba (Qwen)
63.3%source——
21GPT-6.1 Sol Estimated
OpenAI
60.0%source——
22Muse Spark 1.3 Estimated
Meta
58.6%source——
23GPT-5.6 Terra Estimated
OpenAI
58.2%source——
24GLM-5.3 Supported
Z.ai (Zhipu)
56.6%source——
25Grok 4.5 Estimated
xAI
56.2%source——
26Claude Haiku 5.5 Estimated
Anthropic
56.1%source——
27GPT-5.5 Supported
OpenAI
55.1%source——
28Claude Sonnet 5 Supported
Anthropic
54.5%source——
29MiMo-V2.6-Pro
Xiaomi
54.4%source——
30GPT-6 Sol Estimated
OpenAI
54.3%source——
31GLM-5.3-Flash Supported
Z.ai (Zhipu)
52.8%source——
32GPT-5.4 Supported
OpenAI
52.4%source——
33Mistral Large 4
Mistral
51.6%source——
34GPT-5.6 Sol Supported
OpenAI
51.4%source——
35Claude Opus 4.7 Supported
Anthropic
49.2%source——
36Claude Opus 4.8 Supported
Anthropic
48.9%source——
37Beam 1.0
Reflection
48.8%source——
38Kimi K3 Supported
Moonshot AI
47.8%source——
39Qwen3.8 27B
Alibaba (Qwen)
47.5%source——
40DeepSeek V4 Pro 0813
DeepSeek
47.3%source——
41Gemini 3.6 Flash Estimated
Google
46.9%source——
42Inkling 2 Small Estimated
thinkingmachines
46.8%source——
43Claude Opus 4.6 Supported
Anthropic
46.3%source——
44Muse Spark 1.2 Estimated
Meta
45.4%source——
45GLM-5.2 Supported
Z.ai (Zhipu)
45.2%source——
46GPT-6 Luna
OpenAI
44.3%source——
47Claude Sonnet 4.6 Supported
Anthropic
43.0%source——
48GPT-5.6 Luna Estimated
OpenAI
43.0%source——
49GLM-5.1 Estimated
Z.ai (Zhipu)
40.9%source——
50DeepSeek V4.1 Flash Estimated
DeepSeek
39.5%source——
51MiniMax-M3 Estimated
MiniMax
37.7%source——
52Kimi K2.7 Code Estimated
Moonshot AI
37.6%source——
53Gemini 3.1 Pro Preview Supported
Google
35.1%source——
54Inkling Estimated
thinkingmachines
33.8%source——
55Muse Spark Open
Meta
33.4%source——
56Muse Spark 1.1 Estimated
Meta
31.8%source——
57Gemini 3.5 Flash Lite
Google
29.3%source——
58Gemini 3.5 Flash Supported
Google
27.5%source——
59Qwen3.5 397B-A17B
Alibaba (Qwen)
24.9%source——
60Nemotron 3 Ultra
NVIDIA
22.7%source——
61DeepSeek V3.2 Supported
DeepSeek
21.3%source——
62GPT OSS 120B Supported
OpenAI
4.4%source——

For models without a APEX-Agents result, the index estimates one from Epoch’s Capabilities Index using a logistic fit on 58 models (residual SD 0.43 on the logit scale). Estimates are labelled and widen the model’s range.