Evaluation & benchmarks for expert learners
443 resources · page 5 of 14
AI Safety Index: Summer 2026 (Future of Life Institute)
Future of Life Institute AI Safety Index Summer 2026: expert grades of AI companies across safety and security domains.
ForesightSafety Bench
ForesightSafety Bench paper: a frontier-risk evaluation framework covering agentic autonomy, AI-for-science, embodied and catastrophic risks.
FrontierMath
by Epoch AI
Epoch AI FrontierMath: benchmark of unpublished expert-level mathematics problems in tiers, with model results.
GAIA
by GAIA
GAIA organisation page on Hugging Face: benchmark of real-world questions for general AI assistants needing reasoning, browsing and tools.
GAIA leaderboard
by GAIA
GAIA leaderboard Space: scores of AI assistants and agents on the GAIA real-world question benchmark.
GEdit-Bench (StepFun Step1X-Edit)
by stepfun-ai
StepFun Step1X-Edit GitHub repo: open image-editing model whose release includes GEdit-Bench, a benchmark of real-world editing requests.
GenAI-Arena (TIGER Lab)
by TIGER-Lab
TIGER Lab GenAI-Arena Space: academic arena voting between generative image, editing and video models.
GenEval
by Dhruba Ghosh
GenEval GitHub repo: object-focused framework for evaluating text-to-image alignment.
GEO-Bench
by ServiceNow
GEO-Bench GitHub repo from ServiceNow: benchmark for foundation models on Earth-observation classification and segmentation.
Global-MMLU
by Cohere Labs
Cohere Labs Global-MMLU dataset: MMLU translated into many languages with labels marking culturally sensitive questions.
GLUE
by GLUE team (NYU, University of Washington and DeepMind)
GLUE benchmark site: collection of natural-language-understanding tasks with a leaderboard for comparing models.
GPQA
GPQA GitHub repo: graduate-level, Google-proof science questions with baselines for evaluating expert reasoning.
GQA
GQA dataset site: real-image visual question answering with scene graphs for testing compositional reasoning.
GSM8K
by OpenAI
GSM8K GitHub repo from OpenAI: grade-school math word problems with solutions and evaluation code; archived.
GuacaMol
by BenevolentAI
BenevolentAI GuacaMol GitHub repo: benchmarks for generative chemistry models.
Habitat Challenge
Annual Habitat Challenge for embodied AI, with simulated navigation and object-rearrangement tasks, rules and leaderboards for agent evaluation.
HAE-RAE Bench
HAE-RAE Bench GitHub repo: a Korean benchmark on vocabulary, history, general knowledge and reading comprehension.
HAL: Holistic Agent Leaderboard
Princeton HAL Holistic Agent Leaderboard: third-party, cost-aware evaluation of AI agents across benchmarks.
HallusionBench
by Tianyi Lab
HallusionBench GitHub repo (CVPR 2024): benchmark of language hallucination and visual illusion in VLMs.
Harbor framework
by Harbor framework project (Laude Institute)
Harbor framework site: tooling for evaluating and optimising sandboxed agents, running datasets such as Terminal-Bench and SWE-bench Pro.
HarmBench
by Center for AI Safety (HarmBench)
HarmBench site: standardised evaluation framework for automated red-teaming attacks and model refusal defences.
HEAR
HEAR benchmark site: holistic evaluation of audio representations across speech, music and environmental sound tasks.
Hebrew LLM Leaderboard
by Open Hebrew LLM's Leaderboard
Hebrew LLM Leaderboard Hugging Face Space: model results on Hebrew language benchmarks.
HEIM
Stanford CRFM HEIM: holistic evaluation of text-to-image models on quality, alignment, bias and other dimensions.
HELM (Stanford CRFM)
Stanford CRFM HELM: holistic evaluation of language models across many scenarios and metrics.
HELMET
by Princeton Natural Language Processing
HELMET GitHub repo from Princeton: benchmark for long-context language models across seven application categories.
HF Find a Leaderboard
by OpenEvals
Hugging Face Find a Leaderboard Space: searchable directory of leaderboards hosted on the Hub.
Hugging Face LightEval
by Hugging Face
Hugging Face LightEval GitHub repo: toolkit for evaluating LLMs across many backends and tasks.
Humanity's Last Exam
by Center for AI Safety and Scale AI
Humanity's Last Exam: dataset of 2,500 expert-written questions across academic fields, with paper and results.
Humanloop
by Humanloop
Humanloop site, now a notice that the team joined Anthropic and the platform was shut down with a migration guide.
IberBench
by Universitat Politècnica de València
UPV repository record for IberBench, a benchmark for evaluating LLMs on Iberian languages such as Spanish, Catalan and Basque.
IFBench
by Valentina Pyatkin et al. (University of Washington)
IFBench paper on generalising verifiable instruction following, with out-of-distribution output constraints for evaluating LLMs.