{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:CGB34AKJUCDLA4MRT5AZDRXZRR","short_pith_number":"pith:CGB34AKJ","canonical_record":{"source":{"id":"2605.08384","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-08T18:45:15Z","cross_cats_sorted":[],"title_canon_sha256":"9540f9bd214d9de24038e1f3e6692af44f4e1d34e099310e4337c05a2bf1996e","abstract_canon_sha256":"ea6470d86e28138f1baf8a6d4f9cb1cd4938bdb7391a816f931cc60580c14b78"},"schema_version":"1.0"},"canonical_sha256":"1183be0149a086b071919f4191c6f98c6c79a0ad26d4379f806c5e71967b0f24","source":{"kind":"arxiv","id":"2605.08384","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.08384","created_at":"2026-06-09T01:04:43Z"},{"alias_kind":"arxiv_version","alias_value":"2605.08384v3","created_at":"2026-06-09T01:04:43Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.08384","created_at":"2026-06-09T01:04:43Z"},{"alias_kind":"pith_short_12","alias_value":"CGB34AKJUCDL","created_at":"2026-06-09T01:04:43Z"},{"alias_kind":"pith_short_16","alias_value":"CGB34AKJUCDLA4MR","created_at":"2026-06-09T01:04:43Z"},{"alias_kind":"pith_short_8","alias_value":"CGB34AKJ","created_at":"2026-06-09T01:04:43Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:CGB34AKJUCDLA4MRT5AZDRXZRR","target":"record","payload":{"canonical_record":{"source":{"id":"2605.08384","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-08T18:45:15Z","cross_cats_sorted":[],"title_canon_sha256":"9540f9bd214d9de24038e1f3e6692af44f4e1d34e099310e4337c05a2bf1996e","abstract_canon_sha256":"ea6470d86e28138f1baf8a6d4f9cb1cd4938bdb7391a816f931cc60580c14b78"},"schema_version":"1.0"},"canonical_sha256":"1183be0149a086b071919f4191c6f98c6c79a0ad26d4379f806c5e71967b0f24","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T01:04:43.449198Z","signature_b64":"x3cYWkUdLKnPPc/n3eG0vf9jaimQXyt4GGWRGwrrpTrjpMaPo/lwr5Ps9ErG3X5HkgZwSUi8UzzsrigqnGE4Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1183be0149a086b071919f4191c6f98c6c79a0ad26d4379f806c5e71967b0f24","last_reissued_at":"2026-06-09T01:04:43.448778Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T01:04:43.448778Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.08384","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-09T01:04:43Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"l43GQC11bZp9BOT+67nZaGcs05vgdwvm+/AlLXJnSlJ49FhA6Gx5/sTjx0z53Y12hoDDZJoLS2IwkXAmDY58BA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-01T23:22:05.404791Z"},"content_sha256":"6edf48f95e8f96b61dd2a9f75e49ef0605fae866300203d7e79221900c639003","schema_version":"1.0","event_id":"sha256:6edf48f95e8f96b61dd2a9f75e49ef0605fae866300203d7e79221900c639003"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:CGB34AKJUCDLA4MRT5AZDRXZRR","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"jina-embeddings-v5-omni: Geometry-preserving Embeddings via Locked Aligned Towers","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"GELATO extends existing text embedding models to images, audio and video by freezing nearly all weights and training only the connectors.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Andreas Koukounas, Florian H\\\"onicke, Han Xiao, Michael G\\\"unther, Mohammad Kalim Akram, Saba Sturua, Scott Martens","submitted_at":"2026-05-08T18:45:15Z","abstract_excerpt":"In this work, we introduce GELATO (Geometry-preserving Embeddings via Locked Aligned TOwers), a novel approach to multimodal embedding models. We build on the VLM-style architecture, in which non-text encoders are adapted to produce input for a language model, which in turn generates embeddings for all varieties of input. We present the result: the jina-embeddings-v5-omni suite, a pair of models that encode text, image, audio, and video input into a single semantic embedding space. GELATO extends the two Jina Embeddings v5 Text models to support additional modality by adding encoders for image"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our evaluations show that GELATO produces results that are competitive with the state-of-the-art, yielding nearly equal performance to larger multimodal embedding models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that freezing the backbone text embedding models and non-text modality encoders while training only the connecting components will preserve semantic geometry and enable effective cross-modal alignment without degrading original text performance.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GELATO extends frozen text embedding models with locked image and audio encoders, training minimal connectors to produce a single semantic embedding space for text, image, audio, and video while keeping original text performance unchanged.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"GELATO extends existing text embedding models to images, audio and video by freezing nearly all weights and training only the connectors.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"c3594892dd9a4907ae8703229681285853e6f686349dd4e5a4fede4abd2e9dd1"},"source":{"id":"2605.08384","kind":"arxiv","version":3},"verdict":{"id":"dd4f0eb4-5bf3-48cd-b29e-0c9266833e14","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T06:48:58.974981Z","strongest_claim":"Our evaluations show that GELATO produces results that are competitive with the state-of-the-art, yielding nearly equal performance to larger multimodal embedding models.","one_line_summary":"GELATO extends frozen text embedding models with locked image and audio encoders, training minimal connectors to produce a single semantic embedding space for text, image, audio, and video while keeping original text performance unchanged.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that freezing the backbone text embedding models and non-text modality encoders while training only the connecting components will preserve semantic geometry and enable effective cross-modal alignment without degrading original text performance.","pith_extraction_headline":"GELATO extends existing text embedding models to images, audio and video by freezing nearly all weights and training only the connectors."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.08384/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T09:42:04.164772Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-20T04:37:37.335537Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T15:01:18.150795Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T11:10:33.608858Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"234c7c50b95060a113944ced97df1056e83845018dc0d2bd952b09344f98277f"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"7182e2edb58518ea29246fefe9022b65eee71f30e775a5acfb33d6ac6b0030ab"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"dd4f0eb4-5bf3-48cd-b29e-0c9266833e14"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-09T01:04:43Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"KtdMsgPLt4kq3zdOQmOOX4NjyjaLacZ0x4I33XgHhNjaT1wQUw9RwN4Nz1K/3SvuUzszOrwB63kYOrI9vOldBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-01T23:22:05.405329Z"},"content_sha256":"6e6ad682336beecaf6cba1085d4c98c081ee85aebd995cf662ddddc784bf7a36","schema_version":"1.0","event_id":"sha256:6e6ad682336beecaf6cba1085d4c98c081ee85aebd995cf662ddddc784bf7a36"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR/bundle.json","state_url":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/CGB34AKJUCDLA4MRT5AZDRXZRR/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-01T23:22:05Z","links":{"resolver":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR","bundle":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR/bundle.json","state":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR/state.json","well_known_bundle":"https://pith.science/.well-known/pith/CGB34AKJUCDLA4MRT5AZDRXZRR/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:CGB34AKJUCDLA4MRT5AZDRXZRR","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"ea6470d86e28138f1baf8a6d4f9cb1cd4938bdb7391a816f931cc60580c14b78","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-08T18:45:15Z","title_canon_sha256":"9540f9bd214d9de24038e1f3e6692af44f4e1d34e099310e4337c05a2bf1996e"},"schema_version":"1.0","source":{"id":"2605.08384","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.08384","created_at":"2026-06-09T01:04:43Z"},{"alias_kind":"arxiv_version","alias_value":"2605.08384v3","created_at":"2026-06-09T01:04:43Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.08384","created_at":"2026-06-09T01:04:43Z"},{"alias_kind":"pith_short_12","alias_value":"CGB34AKJUCDL","created_at":"2026-06-09T01:04:43Z"},{"alias_kind":"pith_short_16","alias_value":"CGB34AKJUCDLA4MR","created_at":"2026-06-09T01:04:43Z"},{"alias_kind":"pith_short_8","alias_value":"CGB34AKJ","created_at":"2026-06-09T01:04:43Z"}],"graph_snapshots":[{"event_id":"sha256:6e6ad682336beecaf6cba1085d4c98c081ee85aebd995cf662ddddc784bf7a36","target":"graph","created_at":"2026-06-09T01:04:43Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Our evaluations show that GELATO produces results that are competitive with the state-of-the-art, yielding nearly equal performance to larger multimodal embedding models."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The assumption that freezing the backbone text embedding models and non-text modality encoders while training only the connecting components will preserve semantic geometry and enable effective cross-modal alignment without degrading original text performance."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"GELATO extends frozen text embedding models with locked image and audio encoders, training minimal connectors to produce a single semantic embedding space for text, image, audio, and video while keeping original text performance unchanged."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"GELATO extends existing text embedding models to images, audio and video by freezing nearly all weights and training only the connectors."}],"snapshot_sha256":"c3594892dd9a4907ae8703229681285853e6f686349dd4e5a4fede4abd2e9dd1"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"7182e2edb58518ea29246fefe9022b65eee71f30e775a5acfb33d6ac6b0030ab"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T09:42:04.164772Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T04:37:37.335537Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T15:01:18.150795Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T11:10:33.608858Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.08384/integrity.json","findings":[],"snapshot_sha256":"234c7c50b95060a113944ced97df1056e83845018dc0d2bd952b09344f98277f","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"In this work, we introduce GELATO (Geometry-preserving Embeddings via Locked Aligned TOwers), a novel approach to multimodal embedding models. We build on the VLM-style architecture, in which non-text encoders are adapted to produce input for a language model, which in turn generates embeddings for all varieties of input. We present the result: the jina-embeddings-v5-omni suite, a pair of models that encode text, image, audio, and video input into a single semantic embedding space. GELATO extends the two Jina Embeddings v5 Text models to support additional modality by adding encoders for image","authors_text":"Andreas Koukounas, Florian H\\\"onicke, Han Xiao, Michael G\\\"unther, Mohammad Kalim Akram, Saba Sturua, Scott Martens","cross_cats":[],"headline":"GELATO extends existing text embedding models to images, audio and video by freezing nearly all weights and training only the connectors.","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-08T18:45:15Z","title":"jina-embeddings-v5-omni: Geometry-preserving Embeddings via Locked Aligned Towers"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.08384","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-13T06:48:58.974981Z","id":"dd4f0eb4-5bf3-48cd-b29e-0c9266833e14","model_set":{"reader":"grok-4.3"},"one_line_summary":"GELATO extends frozen text embedding models with locked image and audio encoders, training minimal connectors to produce a single semantic embedding space for text, image, audio, and video while keeping original text performance unchanged.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"GELATO extends existing text embedding models to images, audio and video by freezing nearly all weights and training only the connectors.","strongest_claim":"Our evaluations show that GELATO produces results that are competitive with the state-of-the-art, yielding nearly equal performance to larger multimodal embedding models.","weakest_assumption":"The assumption that freezing the backbone text embedding models and non-text modality encoders while training only the connecting components will preserve semantic geometry and enable effective cross-modal alignment without degrading original text performance."}},"verdict_id":"dd4f0eb4-5bf3-48cd-b29e-0c9266833e14"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:6edf48f95e8f96b61dd2a9f75e49ef0605fae866300203d7e79221900c639003","target":"record","created_at":"2026-06-09T01:04:43Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"ea6470d86e28138f1baf8a6d4f9cb1cd4938bdb7391a816f931cc60580c14b78","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-08T18:45:15Z","title_canon_sha256":"9540f9bd214d9de24038e1f3e6692af44f4e1d34e099310e4337c05a2bf1996e"},"schema_version":"1.0","source":{"id":"2605.08384","kind":"arxiv","version":3}},"canonical_sha256":"1183be0149a086b071919f4191c6f98c6c79a0ad26d4379f806c5e71967b0f24","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"1183be0149a086b071919f4191c6f98c6c79a0ad26d4379f806c5e71967b0f24","first_computed_at":"2026-06-09T01:04:43.448778Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-09T01:04:43.448778Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"x3cYWkUdLKnPPc/n3eG0vf9jaimQXyt4GGWRGwrrpTrjpMaPo/lwr5Ps9ErG3X5HkgZwSUi8UzzsrigqnGE4Bg==","signature_status":"signed_v1","signed_at":"2026-06-09T01:04:43.449198Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.08384","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:6edf48f95e8f96b61dd2a9f75e49ef0605fae866300203d7e79221900c639003","sha256:6e6ad682336beecaf6cba1085d4c98c081ee85aebd995cf662ddddc784bf7a36"],"state_sha256":"9433167e53aaa42bd37c5db6c721e9bfcd84dbd5864dc0c8ff3610f3e8c5e5ee"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"JmcmmivBy1VkTjsO7eNAXzBXDAysGrMbpYsUz1+q1CDKpjuPAn/2HnxRKlasyCNbMUlG1eJ53YbuOHyfgaiPAA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-01T23:22:05.408583Z","bundle_sha256":"c6413de0a8bf6f2581e000beb7f247ed55bc69c48c433d9abf283a2bff8ba3a9"}}