Benchmarks & RL envs

Evals26

Eval	Kind	Capability	Scores
AA Long-Context Reasoning	benchmark	Long Context	339
Terminal-Bench Hard	benchmark	Agents	316
τ²-bench	benchmark	Agents	324
IFBench	benchmark	Instruction Following	330
SciCode	benchmark	Code	402
Humanity's Last Exam	benchmark	Reasoning	404
MuSR Multistep soft reasoning	benchmark	Reasoning	1480
MATH Lvl 5 Hardest competition math	benchmark	Math	1480
ARC-AGI Abstraction & reasoning	benchmark	Reasoning	—
MMMU Multimodal understanding	benchmark	Vision & Multimodal	8
IFEval Instruction following	benchmark	Language & Instruction	1488
BIG-Bench Hard Hard reasoning tasks	benchmark	Reasoning	1480
LiveCodeBench Contamination-free coding	benchmark	Code	279
SWE-bench Verified Real GitHub issues	benchmark	Agents	5
MBPP Basic Python problems	benchmark	Code	—
HumanEval Python code synthesis	benchmark	Code	16
AIME Olympiad math	benchmark	Math	209
MATH Competition math	benchmark	Math	192
GSM8K Grade-school math	benchmark	Math	10
GPQA Diamond Graduate-level QA	benchmark	Reasoning	1909
MMLU-Pro Harder multitask QA	benchmark	Knowledge & QA	1758
MMLU Massive Multitask Language Understanding	benchmark	Knowledge & QA	17
BabyAI Grounded instruction env	rl-env	Agents	—
ALFWorld Embodied text env	rl-env	Agents	—
SWE-Gym SWE training environment	rl-env	Agents	—
WebArena Web agent environment	rl-env	Agents	—