Benchmarks

92 AI benchmarks with their saturation status, contamination risk and the source of every score. Filter the list, or scan the trust matrix below to see which capabilities still have benchmarks that separate frontier models.

Capability Domains
Status
Contamination
Sort By
 
Displaying 18 of 18 benchmarks (filtered from 92)
Long ContextNearing Saturation

RULER

Configurable synthetic benchmark that expands needle retrieval into multi-needle, multi-hop tracing, and aggregation tasks.

Trust scoreout of 100
Low contamination
Instruction FollowingNearing Saturation

AgentIF

707 instructions from 50 real-world agents, averaging 1,723 words and 11.9 constraints — whether instruction following survives deployment-scale prompts.

Trust scoreout of 100
Low contamination
Instruction FollowingNearing Saturation

ComplexBench

Bilingual benchmark of how models follow instructions when several constraints must hold at once — where stacking constraints is not linear.

Trust scoreout of 100
Low contamination
KnowledgeNearing Saturation

GPQA

PhD-written science questions so hard that skilled non-experts with Google score 34% — the 'Google-proof' exam, reported on its 198-question Diamond subset.

Trust scoreout of 100
Medium contamination
Long ContextNearing Saturation

ZeroSCROLLS

Zero-shot long-text suite derived from SCROLLS with test-only tasks and new aggregation-style long-context challenges.

Trust scoreout of 100
Medium contamination
Long ContextNearing Saturation

QuALITY

Multiple-choice long-document QA where writers and validators read the full article, with a hard subset designed to defeat skimming.

Trust scoreout of 100
Medium contamination
ReasoningNearing Saturation

ARC-AGI-2

The harder static ARC successor: human-calibrated colored-grid abstraction tasks, now pressured by 2026 frontier systems.

Trust scoreout of 100
Medium contamination
Instruction FollowingNearing Saturation

StructFlowBench

155 multi-turn dialogues scored on fine-grained constraints plus six inter-turn dialogue relationships (recall, refinement, expansion, follow-up, summary).

Trust scoreout of 100
Medium contamination
KnowledgeNearing Saturation

MMLU-Pro

MMLU rebuilt to fix its flaws: 10 options instead of 4, trivia and label noise filtered out, and reasoning-heavy questions that reward chain-of-thought.

Trust scoreout of 100
Medium contamination
MathematicsNearing Saturation

AIME

The MAA's competition exam — 15 integer-answer problems per test, refreshed yearly so each new exam is briefly contamination-free.

Trust scoreout of 100
Medium contamination
Instruction FollowingNearing Saturation

MT-Bench

80 curated multi-turn questions across 8 skills, each a two-turn conversation scored by a strong LLM judge on a 1–10 scale.

Trust scoreout of 100
Medium contamination
Long ContextNearing Saturation

InfiniteBench

100k+ token benchmark mixing realistic and synthetic tasks across English, Chinese, code, math, and retrieval.

Trust scoreout of 100
High contamination
CodingNearing Saturation

MBPP+

A cleaned 378-task MBPP evaluation with roughly 35 times more tests for stricter Python functional correctness.

Trust scoreout of 100
High contamination
Long ContextNearing Saturation

LongBench

Bilingual multitask suite for long-context understanding across QA, summarization, few-shot learning, synthetic tasks, and code completion.

Trust scoreout of 100
High contamination
MathematicsNearing Saturation

MGSM

A human-translated 250-problem GSM8K subset for comparing grade-school mathematical reasoning across ten languages.

Trust scoreout of 100
High contamination
Long ContextNearing Saturation

SCROLLS

Seven-task long-sequence suite covering summarization, QA, and NLI over naturally long English texts.

Trust scoreout of 100
High contamination
Long ContextNearing Saturation

QASPER

Scientific-paper QA over full NLP papers, with abstractive, extractive, yes/no, and unanswerable answers plus supporting evidence.

Trust scoreout of 100
High contamination
Long ContextNearing Saturation

QMSum

Query-focused meeting summarization over long transcripts from academic, product, and committee meetings.

Trust scoreout of 100
High contamination

Trust matrix

Every benchmark by capability and status. Click a benchmark for its evidence and scores.

Contamination Risk
Refresh Cadence
Showing 92 of 92 benchmarks
CapabilityActiveNearing SaturationSaturatedDeprecated
Knowledge
Reasoning
Mathematics
Coding
Instruction Following
Long Context
Safety & Alignment
Multimodal Vision
Agentic & Tool Use