{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:QVKIMRDDQWE4QVQKDLMBEQSV6A","short_pith_number":"pith:QVKIMRDD","canonical_record":{"source":{"id":"2509.24621","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-09-29T11:28:42Z","cross_cats_sorted":[],"title_canon_sha256":"28d5ecb2143721bccfa29937fb3b3e60eb9bd17a79132806c4f7c4ea8a34b242","abstract_canon_sha256":"d1809f56e11bc0656c73999d383998a0d9bc3ea87768ec3f3e351907e262ee69"},"schema_version":"1.0"},"canonical_sha256":"85548644638589c8560a1ad8124255f0042fb6b06f5d33afd6d254834b63acbf","source":{"kind":"arxiv","id":"2509.24621","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2509.24621","created_at":"2026-05-26T02:03:55Z"},{"alias_kind":"arxiv_version","alias_value":"2509.24621v3","created_at":"2026-05-26T02:03:55Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.24621","created_at":"2026-05-26T02:03:55Z"},{"alias_kind":"pith_short_12","alias_value":"QVKIMRDDQWE4","created_at":"2026-05-26T02:03:55Z"},{"alias_kind":"pith_short_16","alias_value":"QVKIMRDDQWE4QVQK","created_at":"2026-05-26T02:03:55Z"},{"alias_kind":"pith_short_8","alias_value":"QVKIMRDD","created_at":"2026-05-26T02:03:55Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:QVKIMRDDQWE4QVQKDLMBEQSV6A","target":"record","payload":{"canonical_record":{"source":{"id":"2509.24621","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-09-29T11:28:42Z","cross_cats_sorted":[],"title_canon_sha256":"28d5ecb2143721bccfa29937fb3b3e60eb9bd17a79132806c4f7c4ea8a34b242","abstract_canon_sha256":"d1809f56e11bc0656c73999d383998a0d9bc3ea87768ec3f3e351907e262ee69"},"schema_version":"1.0"},"canonical_sha256":"85548644638589c8560a1ad8124255f0042fb6b06f5d33afd6d254834b63acbf","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:03:55.741930Z","signature_b64":"+ZyG40iMHx5o1WVfbIeAruvIK3WF3IN/mYNpf9L5+SK9WHFaOkGezkkYVwQEOf2PwobhpFW5BXjjtLxHPxCjBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"85548644638589c8560a1ad8124255f0042fb6b06f5d33afd6d254834b63acbf","last_reissued_at":"2026-05-26T02:03:55.741231Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:03:55.741231Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2509.24621","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":"2026-05-26T02:03:55Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Tn4Hjal2Beu9jV9durImoryYmPExlOvw0aqZgNSmcfQP8LzXjGRJy6WgN0YRkw74hLmN/Lo+MzjmgSSuFJsJDA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-30T19:19:52.324613Z"},"content_sha256":"c3a9f90505bf29b3c6622f713475143651f019668e67c2f8ed0eeadbd850ffcc","schema_version":"1.0","event_id":"sha256:c3a9f90505bf29b3c6622f713475143651f019668e67c2f8ed0eeadbd850ffcc"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:QVKIMRDDQWE4QVQKDLMBEQSV6A","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"FreeRet: MLLMs as Training-Free Retrievers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Off-the-shelf MLLMs can serve as powerful multimodal retrievers without any training by deriving faithful embeddings for search and using reasoning for reranking.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Chenting Wang, Chunxu Liu, Limin Wang, Xiangyu Zeng, Xinhao Li, Yicheng Xu, Yi Wang, Yuhan Zhu, Ziang Yan","submitted_at":"2025-09-29T11:28:42Z","abstract_excerpt":"Multimodal large language models (MLLMs) are emerging as versatile foundations for mixed-modality retrieval. Yet, they often require heavy post-hoc training to convert them into contrastive encoders for retrieval. This work asks: Can off-the-shelf MLLMs serve as powerful retrievers without additional training? We present FreeRet, a plug-and-play framework that turns any MLLM into a two-stage retriever. FreeRet first derives semantically grounded embeddings directly from the model for fast candidate search, and then exploits its reasoning ability for precise reranking. The framework contributes"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On the MMEB and MMEB-V2 benchmarks spanning 46 datasets, FreeRet substantially outperforms models trained on millions of pairs.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That off-the-shelf MLLMs already contain semantically faithful embeddings and reliable reasoning capabilities that can be directly harnessed for retrieval without any post-hoc training or alignment adjustments.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"FreeRet enables pretrained MLLMs to act as training-free retrievers via semantically grounded embeddings and reasoning-based reranking, outperforming models trained on millions of pairs on MMEB benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Off-the-shelf MLLMs can serve as powerful multimodal retrievers without any training by deriving faithful embeddings for search and using reasoning for reranking.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"bad4a03c81e9fa3a402c6f459a1f84567f23e6f6d9c52732249d6700b923d28e"},"source":{"id":"2509.24621","kind":"arxiv","version":3},"verdict":{"id":"218d5d32-48f0-44a2-bec3-7844f329492a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-18T12:56:48.731362Z","strongest_claim":"On the MMEB and MMEB-V2 benchmarks spanning 46 datasets, FreeRet substantially outperforms models trained on millions of pairs.","one_line_summary":"FreeRet enables pretrained MLLMs to act as training-free retrievers via semantically grounded embeddings and reasoning-based reranking, outperforming models trained on millions of pairs on MMEB benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That off-the-shelf MLLMs already contain semantically faithful embeddings and reliable reasoning capabilities that can be directly harnessed for retrieval without any post-hoc training or alignment adjustments.","pith_extraction_headline":"Off-the-shelf MLLMs can serve as powerful multimodal retrievers without any training by deriving faithful embeddings for search and using reasoning for reranking."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2509.24621/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":2,"snapshot_sha256":"c2f8b9fc5a4b91e3f867ddadc96bc060fea7999ee2aac06ad61060ecd34ba522"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"218d5d32-48f0-44a2-bec3-7844f329492a"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-26T02:03:55Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"G4bTZly65xg4dtVY+k+483vx0EmJYpzmp+D7NzuG5ZxzfKsjFPsEB1p52/kZuzhGzdfsIExITZEoGsnXJtOkDQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-30T19:19:52.325111Z"},"content_sha256":"4e6554105c019526756c5ad29356636fd8da491e9f5eb3c7fd99e4ccb80ee597","schema_version":"1.0","event_id":"sha256:4e6554105c019526756c5ad29356636fd8da491e9f5eb3c7fd99e4ccb80ee597"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/QVKIMRDDQWE4QVQKDLMBEQSV6A/bundle.json","state_url":"https://pith.science/pith/QVKIMRDDQWE4QVQKDLMBEQSV6A/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/QVKIMRDDQWE4QVQKDLMBEQSV6A/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":"2026-05-30T19:19:52Z","links":{"resolver":"https://pith.science/pith/QVKIMRDDQWE4QVQKDLMBEQSV6A","bundle":"https://pith.science/pith/QVKIMRDDQWE4QVQKDLMBEQSV6A/bundle.json","state":"https://pith.science/pith/QVKIMRDDQWE4QVQKDLMBEQSV6A/state.json","well_known_bundle":"https://pith.science/.well-known/pith/QVKIMRDDQWE4QVQKDLMBEQSV6A/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:QVKIMRDDQWE4QVQKDLMBEQSV6A","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":"d1809f56e11bc0656c73999d383998a0d9bc3ea87768ec3f3e351907e262ee69","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-09-29T11:28:42Z","title_canon_sha256":"28d5ecb2143721bccfa29937fb3b3e60eb9bd17a79132806c4f7c4ea8a34b242"},"schema_version":"1.0","source":{"id":"2509.24621","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2509.24621","created_at":"2026-05-26T02:03:55Z"},{"alias_kind":"arxiv_version","alias_value":"2509.24621v3","created_at":"2026-05-26T02:03:55Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.24621","created_at":"2026-05-26T02:03:55Z"},{"alias_kind":"pith_short_12","alias_value":"QVKIMRDDQWE4","created_at":"2026-05-26T02:03:55Z"},{"alias_kind":"pith_short_16","alias_value":"QVKIMRDDQWE4QVQK","created_at":"2026-05-26T02:03:55Z"},{"alias_kind":"pith_short_8","alias_value":"QVKIMRDD","created_at":"2026-05-26T02:03:55Z"}],"graph_snapshots":[{"event_id":"sha256:4e6554105c019526756c5ad29356636fd8da491e9f5eb3c7fd99e4ccb80ee597","target":"graph","created_at":"2026-05-26T02:03:55Z","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":"On the MMEB and MMEB-V2 benchmarks spanning 46 datasets, FreeRet substantially outperforms models trained on millions of pairs."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That off-the-shelf MLLMs already contain semantically faithful embeddings and reliable reasoning capabilities that can be directly harnessed for retrieval without any post-hoc training or alignment adjustments."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"FreeRet enables pretrained MLLMs to act as training-free retrievers via semantically grounded embeddings and reasoning-based reranking, outperforming models trained on millions of pairs on MMEB benchmarks."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Off-the-shelf MLLMs can serve as powerful multimodal retrievers without any training by deriving faithful embeddings for search and using reasoning for reranking."}],"snapshot_sha256":"bad4a03c81e9fa3a402c6f459a1f84567f23e6f6d9c52732249d6700b923d28e"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"c2f8b9fc5a4b91e3f867ddadc96bc060fea7999ee2aac06ad61060ecd34ba522"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2509.24621/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Multimodal large language models (MLLMs) are emerging as versatile foundations for mixed-modality retrieval. Yet, they often require heavy post-hoc training to convert them into contrastive encoders for retrieval. This work asks: Can off-the-shelf MLLMs serve as powerful retrievers without additional training? We present FreeRet, a plug-and-play framework that turns any MLLM into a two-stage retriever. FreeRet first derives semantically grounded embeddings directly from the model for fast candidate search, and then exploits its reasoning ability for precise reranking. The framework contributes","authors_text":"Chenting Wang, Chunxu Liu, Limin Wang, Xiangyu Zeng, Xinhao Li, Yicheng Xu, Yi Wang, Yuhan Zhu, Ziang Yan","cross_cats":[],"headline":"Off-the-shelf MLLMs can serve as powerful multimodal retrievers without any training by deriving faithful embeddings for search and using reasoning for reranking.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-09-29T11:28:42Z","title":"FreeRet: MLLMs as Training-Free Retrievers"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.24621","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-18T12:56:48.731362Z","id":"218d5d32-48f0-44a2-bec3-7844f329492a","model_set":{"reader":"grok-4.3"},"one_line_summary":"FreeRet enables pretrained MLLMs to act as training-free retrievers via semantically grounded embeddings and reasoning-based reranking, outperforming models trained on millions of pairs on MMEB benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Off-the-shelf MLLMs can serve as powerful multimodal retrievers without any training by deriving faithful embeddings for search and using reasoning for reranking.","strongest_claim":"On the MMEB and MMEB-V2 benchmarks spanning 46 datasets, FreeRet substantially outperforms models trained on millions of pairs.","weakest_assumption":"That off-the-shelf MLLMs already contain semantically faithful embeddings and reliable reasoning capabilities that can be directly harnessed for retrieval without any post-hoc training or alignment adjustments."}},"verdict_id":"218d5d32-48f0-44a2-bec3-7844f329492a"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:c3a9f90505bf29b3c6622f713475143651f019668e67c2f8ed0eeadbd850ffcc","target":"record","created_at":"2026-05-26T02:03:55Z","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":"d1809f56e11bc0656c73999d383998a0d9bc3ea87768ec3f3e351907e262ee69","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-09-29T11:28:42Z","title_canon_sha256":"28d5ecb2143721bccfa29937fb3b3e60eb9bd17a79132806c4f7c4ea8a34b242"},"schema_version":"1.0","source":{"id":"2509.24621","kind":"arxiv","version":3}},"canonical_sha256":"85548644638589c8560a1ad8124255f0042fb6b06f5d33afd6d254834b63acbf","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"85548644638589c8560a1ad8124255f0042fb6b06f5d33afd6d254834b63acbf","first_computed_at":"2026-05-26T02:03:55.741231Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-26T02:03:55.741231Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"+ZyG40iMHx5o1WVfbIeAruvIK3WF3IN/mYNpf9L5+SK9WHFaOkGezkkYVwQEOf2PwobhpFW5BXjjtLxHPxCjBw==","signature_status":"signed_v1","signed_at":"2026-05-26T02:03:55.741930Z","signed_message":"canonical_sha256_bytes"},"source_id":"2509.24621","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:c3a9f90505bf29b3c6622f713475143651f019668e67c2f8ed0eeadbd850ffcc","sha256:4e6554105c019526756c5ad29356636fd8da491e9f5eb3c7fd99e4ccb80ee597"],"state_sha256":"d8f73d89bdfd07a2391c0b20cf60fc6f3dbf7e9d2f2beee267c336948e499e72"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"gZMgsJUPVaUppxou9dF5maRZBCGZMKpnbNIKWkul9K18KH3MC9EgP6uyi/WOb9keMqQlLEBZYSobAA/7f5rKAw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-30T19:19:52.328145Z","bundle_sha256":"4471c091dd0efb68713c623ecd03dc431336a999c7740881e88de15ce16896cd"}}