{"benchmark_id":"big-bench-extra-hard","name":"BIG-Bench Extra Hard","parent_benchmark":null,"categories":["reasoning","language","general"],"modality":"text","multilingual":false,"max_score":1.0,"language":"en","description":"BIG-Bench Extra Hard (BBEH) is a challenging benchmark that replaces each task in BIG-Bench Hard with a novel task that probes similar reasoning capabilities but exhibits significantly increased difficulty. The benchmark contains 23 tasks testing diverse reasoning skills including many-hop reasoning, causal understanding, spatial reasoning, temporal arithmetic, geometric reasoning, linguistic reasoning, logic puzzles, and humor understanding. Designed to address saturation on existing benchmarks where state-of-the-art models achieve near-perfect scores, BBEH shows substantial room for improvement with best models achieving only 9.8-44.8% average accuracy.","paper_link":"https://arxiv.org/abs/2502.19187","implementation_link":null,"verified":false,"created_at":"2026-05-07T16:53:21.974082+00:00","updated_at":"2026-07-05T18:27:44.180892+00:00","statistics":{"total_models":11,"average_score":0.33054545454545453,"min_score":0.072,"max_score":0.744,"score_stddev":0.23241702331643593,"verified_count":0,"self_reported_count":11},"child_benchmarks":[],"linked_dataset":null,"models":[{"rank":1,"model_id":"gemma-4-31b-it","model_name":"Gemma 4 31B","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.744,"normalized_score":0.744,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/blog/gemma4","analysis_method":null,"verification_date":null,"provider_id":"deepinfra","input_cost_per_million":0.13,"output_cost_per_million":0.38,"context_window":262144,"announcement_date":"2026-04-02","param_count":30700000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"DeepInfra","best_throughput":null,"throughput_provider":"DeepInfra","context_provider":"Together"},{"rank":2,"model_id":"gemma-4-26b-a4b-it","model_name":"Gemma 4 26B-A4B","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.648,"normalized_score":0.648,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/blog/gemma4","analysis_method":null,"verification_date":null,"provider_id":"novita","input_cost_per_million":0.13,"output_cost_per_million":0.4,"context_window":262144,"announcement_date":"2026-04-02","param_count":25200000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"Novita","best_throughput":null,"throughput_provider":"Novita","context_provider":"Novita"},{"rank":3,"model_id":"gemma-4-12b-it","model_name":"Gemma 4 12B","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.53,"normalized_score":0.53,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/google/gemma-4-12B-it","analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2026-05-23","param_count":11959730224,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":4,"model_id":"diffusiongemma-26b-a4b-it","model_name":"DiffusionGemma 26B-A4B","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.476,"normalized_score":0.476,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/google/diffusiongemma-26B-A4B-it","analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2026-06-10","param_count":25200000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":5,"model_id":"gemma-4-e4b-it","model_name":"Gemma 4 E4B","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.331,"normalized_score":0.331,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/blog/gemma4","analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2026-04-02","param_count":8000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":6,"model_id":"gemma-4-e2b-it","model_name":"Gemma 4 E2B","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.219,"normalized_score":0.219,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/blog/gemma4","analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2026-04-02","param_count":5100000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":7,"model_id":"gemma-3-27b-it","model_name":"Gemma 3 27B","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.193,"normalized_score":0.193,"verified":false,"self_reported":true,"self_reported_source":"https://ai.google.dev/gemma/docs/core/model_card_3","analysis_method":"0-shot evaluation","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-03-12","param_count":27000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":8,"model_id":"gemma-3-12b-it","model_name":"Gemma 3 12B","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.163,"normalized_score":0.163,"verified":false,"self_reported":true,"self_reported_source":"https://ai.google.dev/gemma/docs/core/model_card_3","analysis_method":"0-shot evaluation","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-03-12","param_count":12000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":9,"model_id":"gemini-diffusion","model_name":"Gemini Diffusion","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.15,"normalized_score":0.15,"verified":false,"self_reported":true,"self_reported_source":"https://deepmind.google/models/gemini-diffusion/","analysis_method":"pass @1","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-05-20","param_count":null,"is_open_source":false,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":10,"model_id":"gemma-3-4b-it","model_name":"Gemma 3 4B","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.11,"normalized_score":0.11,"verified":false,"self_reported":true,"self_reported_source":"https://ai.google.dev/gemma/docs/core/model_card_3","analysis_method":"0-shot evaluation","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-03-12","param_count":4000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":11,"model_id":"gemma-3-1b-it","model_name":"Gemma 3 1B","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.072,"normalized_score":0.072,"verified":false,"self_reported":true,"self_reported_source":"https://ai.google.dev/gemma/docs/core/model_card_3","analysis_method":"0-shot evaluation","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-03-12","param_count":1000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null}]}