DAYJOB: Law Can AI do a lawyer’s job?

Enterprise Instruction Following

ComplexConstraints

A benchmark for professional instruction following, where constraints depend on each other, fire conditionally, and must be inferred from context.

RL Environments and the Hierarchy of Agentic Capabilities
Our RL environment run on 9 models revealed the core capabilities all agents need to master: tool use, planning, adaptability, groundedness, and common sense.
ComplexConstraints Leaderboard
1
Gemini
4 Hydrogen (xHigh reasoning)
73.7
%
1
Gemini
4 Neon (High reasoning)
72.9
%
1
Gemini
4 Carbon (High reasoning)
70.9
%
1
GPT
6 Astra (Max reasoning)
70.8
%
1
Gemini
4 Argon (High reasoning)
70.3
%
1
GPT
6.1 Sol (Max reasoning)
70.1
%
1
GPT
6.2 Astra (Max reasoning)
69.6
%
1
GPT
6.2 Sol (Max reasoning)
69.5
%
1
Muse
Wonder 1.0 (Max reasoning)
67.9
%
1
Claude
Sonnet 5.5 (Adaptive/Max)
67.2
%
1
Hy
Hy4 (High reasoning)
66.8
%
1
Muse
Spark 1.3 (Max reasoning)
64.8
%
1
Claude
Fable 5.2 (Adaptive/Max)
64.5
%
1
Claude
Opus 5.5 (Adaptive/Max)
64.4
%
1
Gemini
3.8 Flash (High reasoning)
63.2
%
1
Grok
4.7 (xHigh reasoning)
60.8
%
1
Claude
Fable 5.1 (Adaptive/Max)
60.4
%
1
Muse
Ember 1.0 (Max reasoning)
58.8
%
1
GLM
5.3 (Max reasoning)
54.4
%
1
Kimi
K3 (Max reasoning)
51.1
%
1
Hy
Hy4 Preview (High reasoning)
49.2
%
1
Inkling
2 Small (xHigh reasoning)
47.2
%
1
Qwen
3.8 Max (xHigh reasoning)
46.3
%
1
Mistral
Large 4 (High reasoning)
42.4
%
1
MiniMax
M3.1 Flash (Max reasoning)
30.9
%

Stay Posted