{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:CGB34AKJUCDLA4MRT5AZDRXZRR","short_pith_number":"pith:CGB34AKJ","schema_version":"1.0","canonical_sha256":"1183be0149a086b071919f4191c6f98c6c79a0ad26d4379f806c5e71967b0f24","source":{"kind":"arxiv","id":"2605.08384","version":3},"attestation_state":"computed","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"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":true},"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"},"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"},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.08384","created_at":"2026-06-09T01:04:43.448835+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.08384v3","created_at":"2026-06-09T01:04:43.448835+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.08384","created_at":"2026-06-09T01:04:43.448835+00:00"},{"alias_kind":"pith_short_12","alias_value":"CGB34AKJUCDL","created_at":"2026-06-09T01:04:43.448835+00:00"},{"alias_kind":"pith_short_16","alias_value":"CGB34AKJUCDLA4MR","created_at":"2026-06-09T01:04:43.448835+00:00"},{"alias_kind":"pith_short_8","alias_value":"CGB34AKJ","created_at":"2026-06-09T01:04:43.448835+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2606.30682","citing_title":"ALM2Vec: Learning Audio Embeddings for Universal Audio Retrieval with Large Audio-Language Models","ref_index":27,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR","json":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR.json","graph_json":"https://pith.science/api/pith-number/CGB34AKJUCDLA4MRT5AZDRXZRR/graph.json","events_json":"https://pith.science/api/pith-number/CGB34AKJUCDLA4MRT5AZDRXZRR/events.json","paper":"https://pith.science/paper/CGB34AKJ"},"agent_actions":{"view_html":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR","download_json":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR.json","view_paper":"https://pith.science/paper/CGB34AKJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.08384&json=true","fetch_graph":"https://pith.science/api/pith-number/CGB34AKJUCDLA4MRT5AZDRXZRR/graph.json","fetch_events":"https://pith.science/api/pith-number/CGB34AKJUCDLA4MRT5AZDRXZRR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR/action/storage_attestation","attest_author":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR/action/author_attestation","sign_citation":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR/action/citation_signature","submit_replication":"https://pith.science/pith/CGB34AKJUCDLA4MRT5AZDRXZRR/action/replication_record"}},"created_at":"2026-06-09T01:04:43.448835+00:00","updated_at":"2026-06-09T01:04:43.448835+00:00"}