{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HXMSTDT75FZM4VLXRJO7YZTEAH","short_pith_number":"pith:HXMSTDT7","schema_version":"1.0","canonical_sha256":"3dd9298e7fe972ce55778a5dfc666401f467befb016892f44b646681bd966b10","source":{"kind":"arxiv","id":"2607.28640","version":1},"attestation_state":"computed","paper":{"title":"TokenSwap: Benchmarking and Reducing the Modality Gap in Multimodal LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Aleksandra Faust, Andong Hua, Arian Hosseini, Colton Bishop, Igor Mordatch, Jindong Gu, Rebecca Roelofs, Yao Qin","submitted_at":"2026-05-20T00:25:52Z","abstract_excerpt":"Multimodal large language models (MLLMs) should generate consistent responses given semantically equivalent inputs across modalities. However, we observe a systematic discrepancy in model predictions under such cross-modal variations. Specifically, we define the modality gap as the difference in model performance under semantically equivalent textual and multimodal inputs. We introduce TokenSwap, a method that constructs such inputs by replacing textual concepts with semantically aligned images, resulting in sequences where visual tokens are interleaved with text tokens. Based on TokenSwap, we"},"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":false},"canonical_record":{"source":{"id":"2607.28640","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-20T00:25:52Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"36564daa8fd7c2252b55fca36aa59a8b9a7e0f16fade69534973db7d3d6a63b4","abstract_canon_sha256":"cfc00a069f1c503882c263f3d3d489a94f3db5df9f5f67e8158200df12cfdee1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-03T00:11:52.883081Z","signature_b64":"4pPwlq2rErvLiUNrfNpCo166T+8/9miCTo1iuOMhLABHkAIqZkTWZtFCbmVeM1VtaHXDer818tVuhMH/Cws5BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3dd9298e7fe972ce55778a5dfc666401f467befb016892f44b646681bd966b10","last_reissued_at":"2026-08-03T00:11:52.880902Z","signature_status":"signed_v1","first_computed_at":"2026-08-03T00:11:52.880902Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TokenSwap: Benchmarking and Reducing the Modality Gap in Multimodal LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Aleksandra Faust, Andong Hua, Arian Hosseini, Colton Bishop, Igor Mordatch, Jindong Gu, Rebecca Roelofs, Yao Qin","submitted_at":"2026-05-20T00:25:52Z","abstract_excerpt":"Multimodal large language models (MLLMs) should generate consistent responses given semantically equivalent inputs across modalities. However, we observe a systematic discrepancy in model predictions under such cross-modal variations. Specifically, we define the modality gap as the difference in model performance under semantically equivalent textual and multimodal inputs. We introduce TokenSwap, a method that constructs such inputs by replacing textual concepts with semantically aligned images, resulting in sequences where visual tokens are interleaved with text tokens. Based on TokenSwap, we"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.28640","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2607.28640/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"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":"2607.28640","created_at":"2026-08-03T00:11:52.882395+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.28640v1","created_at":"2026-08-03T00:11:52.882395+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.28640","created_at":"2026-08-03T00:11:52.882395+00:00"},{"alias_kind":"pith_short_12","alias_value":"HXMSTDT75FZM","created_at":"2026-08-03T00:11:52.882395+00:00"},{"alias_kind":"pith_short_16","alias_value":"HXMSTDT75FZM4VLX","created_at":"2026-08-03T00:11:52.882395+00:00"},{"alias_kind":"pith_short_8","alias_value":"HXMSTDT7","created_at":"2026-08-03T00:11:52.882395+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HXMSTDT75FZM4VLXRJO7YZTEAH","json":"https://pith.science/pith/HXMSTDT75FZM4VLXRJO7YZTEAH.json","graph_json":"https://pith.science/api/pith-number/HXMSTDT75FZM4VLXRJO7YZTEAH/graph.json","events_json":"https://pith.science/api/pith-number/HXMSTDT75FZM4VLXRJO7YZTEAH/events.json","paper":"https://pith.science/paper/HXMSTDT7"},"agent_actions":{"view_html":"https://pith.science/pith/HXMSTDT75FZM4VLXRJO7YZTEAH","download_json":"https://pith.science/pith/HXMSTDT75FZM4VLXRJO7YZTEAH.json","view_paper":"https://pith.science/paper/HXMSTDT7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.28640&json=true","fetch_graph":"https://pith.science/api/pith-number/HXMSTDT75FZM4VLXRJO7YZTEAH/graph.json","fetch_events":"https://pith.science/api/pith-number/HXMSTDT75FZM4VLXRJO7YZTEAH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HXMSTDT75FZM4VLXRJO7YZTEAH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HXMSTDT75FZM4VLXRJO7YZTEAH/action/storage_attestation","attest_author":"https://pith.science/pith/HXMSTDT75FZM4VLXRJO7YZTEAH/action/author_attestation","sign_citation":"https://pith.science/pith/HXMSTDT75FZM4VLXRJO7YZTEAH/action/citation_signature","submit_replication":"https://pith.science/pith/HXMSTDT75FZM4VLXRJO7YZTEAH/action/replication_record"}},"created_at":"2026-08-03T00:11:52.882395+00:00","updated_at":"2026-08-03T00:11:52.882395+00:00"}