{"record": {"aliases": [], "artifacts": [], "categories": [], "description": {}, "documents": [{"document_type": "registry_page", "id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "title": "Belebele"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "benchmark-saturation:44", "modality": null, "name": "Belebele", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"crawl_bundle": "benchmark_saturation_2026-11-04", "crawled_at": "2026-11-04T06:47:00Z", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv"}, "publisher": null, "released": "2024-07-25", "released_reference": {"basis": "benchmark_release", "source_key": "benchmark-saturation:44", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv"}, "schema_version": 1, "sizes": [], "slug": "benchmark-saturation-44", "source": "benchmark_saturation", "source_benchmark_id": "44", "source_metadata": {"benchmark_saturation": {"dataset_issues": "No known issues", "saturation_index": "0.5127354214", "saturation_index_interpreted": "Moderate", "test_samples": "109,800"}}}, "schema_version": 1, "scores_by_source": {"benchmark_saturation": {"rows": [{"comparable_group": null, "crawled_at": "2026-11-04T06:47:00Z", "date_precision": "model_announcement", "document_id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "key": "benchmark-saturation:44", "measured_by": null, "model_id": null, "model_name": "cogito-v1-preview-qwen-14B", "obs_id": "benchmark_saturation:44:unidentified-71e018144041f2baef01", "organization": null, "rank_in_source_response": 1, "raw_value": "88.4", "reported_by": "unknown", "reported_date": null, "series_id": "benchmark_saturation:44:default", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "value": 88.4, "value_kind": "number"}, {"comparable_group": null, "crawled_at": "2026-11-04T06:47:00Z", "date_precision": "model_announcement", "document_id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "key": "benchmark-saturation:44", "measured_by": null, "model_id": null, "model_name": "gemma-3-12b-it,bos", "obs_id": "benchmark_saturation:44:unidentified-bc7f49241fd8761e3122", "organization": null, "rank_in_source_response": 2, "raw_value": "88.2", "reported_by": "unknown", "reported_date": null, "series_id": "benchmark_saturation:44:default", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "value": 88.2, "value_kind": "number"}, {"comparable_group": null, "crawled_at": "2026-11-04T06:47:00Z", "date_precision": "model_announcement", "document_id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "key": "benchmark-saturation:44", "measured_by": null, "model_id": null, "model_name": "Qwen3-14B", "obs_id": "benchmark_saturation:44:unidentified-c7bcc9f5ebbe528e205e", "organization": null, "rank_in_source_response": 3, "raw_value": "87.39", "reported_by": "unknown", "reported_date": null, "series_id": "benchmark_saturation:44:default", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "value": 87.39, "value_kind": "number"}, {"comparable_group": null, "crawled_at": "2026-11-04T06:47:00Z", "date_precision": "model_announcement", "document_id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "key": "benchmark-saturation:44", "measured_by": null, "model_id": null, "model_name": "gemma-2-9b-it", "obs_id": "benchmark_saturation:44:unidentified-a14886e988a84c747f51", "organization": null, "rank_in_source_response": 4, "raw_value": "87.31", "reported_by": "unknown", "reported_date": null, "series_id": "benchmark_saturation:44:default", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "value": 87.31, "value_kind": "number"}, {"comparable_group": null, "crawled_at": "2026-11-04T06:47:00Z", "date_precision": "model_announcement", "document_id": "benchmark_saturation:https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "key": "benchmark-saturation:44", "measured_by": null, "model_id": null, "model_name": "GaMS3-12B-Instruct,bos", "obs_id": "benchmark_saturation:44:unidentified-ac6dff9e854d1cddddb2", "organization": null, "rank_in_source_response": 5, "raw_value": "86.29", "reported_by": "unknown", "reported_date": null, "series_id": "benchmark_saturation:44:default", "source": "benchmark_saturation", "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv", "value": 86.29, "value_kind": "number"}], "series": {"bounds": {"basis": "aggregator_declared", "max": null, "min": null}, "declared_max": null, "direction": "higher_is_better", "direction_basis": "source_rank_descending", "display_scale": null, "first_score_record": null, "first_score_report": null, "key": "benchmark-saturation:44", "max_score_contradicted": false, "metric": "Accuracy", "observation_count": 5, "observed_max": 88.4, "score_summary": {"display_max": 88.4, "display_multiplier": 1, "model_count": null, "model_count_basis": null, "numeric_count": 5, "raw_max": 88.4, "raw_min": 86.29, "source_reference": {"crawled_at": "2026-11-04T06:47:00Z", "obs_id": "benchmark_saturation:44:unidentified-71e018144041f2baef01", "reported_date": null, "source_url": "https://github.com/evaleval/benchmark-saturation/blob/6285e3dda145f575a0ccb3f7f4d70cd429832add/data/manual_annotation_data.csv"}, "unit": null}, "series_id": "benchmark_saturation:44:default"}}}, "siblings": []}