AGI·EVALSSign in
24 open benchmarks live — pip install agi-eval

Plug in.
Run the
eval.

Every major AGI benchmark behind two clean protocols. Bring any model endpoint and score it for real — no directory pretending to be a platform.

FIG. A — EVAL RUN, SIDE ELEVATIONscale 1:1 · units norm.
$ agi-evals run gpqa-diamondopenai:gpt-x
score0.712
✓ pushed · rank 03WRONG_ANSWER 41 · NO_ANSWER 3
project
Benchmark registry
sheet · rev
AE-001 · 0.1
45
evals catalogued
24
runnable today
6
categories
8
model adapters
Fig. B — The architecture

Two protocols

Everything hangs off two interfaces. An eval never imports an adapter; an adapter never imports an eval. Get these right and adding eval #2 through #50 is incremental work.

PatientAdapter

A model endpoint. Takes a prompt plus any eval-specific scenario, returns a response. Ships for every major backend.

def respond(self, request:
        PatientRequest) -> PatientResponse: ...
EvalRunner

An eval. Takes a patient and a case, returns a scored result with a typed failure tag. The harness handles concurrency and errors.

def run_case(self, patient, case:
        Case) -> ScoredResult: ...
Fig. C — Runnable today

Live evals

All 45 evals →
EvalCategoryGradingStatus
GPQA DiamondReasoningmcqLive
MMLU-ProReasoningmcqLive
MATHReasoningmathLive
AIME 2024ReasoningmathLive
BIG-Bench HardReasoningexact-matchLive
MuSRReasoningmcqLive
ZebraLogicReasoninglogicLive
HumanEval+CodecodeLive
LiveCodeBenchCodecodeLive
BigCodeBenchCodecodeLive
RepoBenchCodecodeLive
SWE-LancerCodecodeLive
τ-benchAgent / Tool usetool-useLive
SWE-bench VerifiedAgent / Tool useagentLive
SWE-bench LiteAgent / Tool useagentLive
GAIAAgent / Tool useagentLive
Berkeley Function-Calling LeaderboardAgent / Tool usefunction-callingLive
ALFWorldEmbodiedtext-gameLive
ScienceWorldEmbodiedtext-gameLive
WebShopEmbodiedembodiedLive
LIBERORoboticsroboticsLive
HarmBenchSafety / SecuritysafetyLive
AILuminateSafety / SecuritysafetyLive
JailbreakBenchSafety / SecuritysafetyLive
Fig. E — Leaderboard, exploded view

Climb it. Push it.

Full board →

Run an eval, push your score, watch it over time — then forward your scoreboard to a larger challenge in one click.

#ModelOrgEvalScore
01claude-opus-4.8anthropicgpqa-diamond0.811
02gpt-xopenaigpqa-diamond0.774
03grok-4xaigpqa-diamond0.738
04llama-4-405bmetahumaneval-plus0.901
05qwen3-72balibabamath0.689
Fig. F — Adapters

Eight adapters

OpenAIAnthropicGrokOllamavLLMHF TransformersMLXCustom callable
Fig. G — Provenance

Benchmarks are research.
We credit it like research.

Every eval on this platform is years of someone's work, and scores tend to travel without the names attached. So crediting is built in: citations at the point of use, one-click BibTeX, verified links to star the upstream repo, and public run counts that show authors their work mattering.

How we credit researchers →
specimen
$ agi-evals cite tau-bench
@misc{yao2024taubench,
  title = {τ-bench: A Benchmark for ...},
  author = {Yao and others}, ...
}
★ star upstream → sierra-research/tau-bench

Stop guessing. Measure.