{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:VRNYWBAP4OGGSTHSNUUWSTB2UH","short_pith_number":"pith:VRNYWBAP","canonical_record":{"source":{"id":"2604.04917","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-06T17:56:25Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"b239fcf64cf625c6e1202214cd9c444a5428698b2729a6af594559fb0672bc09","abstract_canon_sha256":"02f3aeeaf574c5c19b0719a3eda8ffe161688a6249952d286b9a3b739270eeeb"},"schema_version":"1.0"},"canonical_sha256":"ac5b8b040fe38c694cf26d29694c3aa1c040180d3d5e4a74933df1554b7f149f","source":{"kind":"arxiv","id":"2604.04917","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.04917","created_at":"2026-06-19T16:12:53Z"},{"alias_kind":"arxiv_version","alias_value":"2604.04917v3","created_at":"2026-06-19T16:12:53Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.04917","created_at":"2026-06-19T16:12:53Z"},{"alias_kind":"pith_short_12","alias_value":"VRNYWBAP4OGG","created_at":"2026-06-19T16:12:53Z"},{"alias_kind":"pith_short_16","alias_value":"VRNYWBAP4OGGSTHS","created_at":"2026-06-19T16:12:53Z"},{"alias_kind":"pith_short_8","alias_value":"VRNYWBAP","created_at":"2026-06-19T16:12:53Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:VRNYWBAP4OGGSTHSNUUWSTB2UH","target":"record","payload":{"canonical_record":{"source":{"id":"2604.04917","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-06T17:56:25Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"b239fcf64cf625c6e1202214cd9c444a5428698b2729a6af594559fb0672bc09","abstract_canon_sha256":"02f3aeeaf574c5c19b0719a3eda8ffe161688a6249952d286b9a3b739270eeeb"},"schema_version":"1.0"},"canonical_sha256":"ac5b8b040fe38c694cf26d29694c3aa1c040180d3d5e4a74933df1554b7f149f","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:12:53.850520Z","signature_b64":"nOeawuoq3LFdme++jrefQimTIVOTUh2++RJrQEvHuqrYJui5Oexq+ADQv4KgMFgWucsLwd827/PH0VnhJVaXCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ac5b8b040fe38c694cf26d29694c3aa1c040180d3d5e4a74933df1554b7f149f","last_reissued_at":"2026-06-19T16:12:53.850107Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:12:53.850107Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2604.04917","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":"2026-06-19T16:12:53Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"tsnyJY9KPqZT4L4Q53tXJYCjaWbCk7CYlq67kUv6Puhp/jjkh04v7SjubB3ERBhrLOZL8Mt7nDfbts764jhWAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-30T14:00:19.292877Z"},"content_sha256":"f5e23062947a9c15645a5c388c353899136b8b211aa03aa12e3f00e045029082","schema_version":"1.0","event_id":"sha256:f5e23062947a9c15645a5c388c353899136b8b211aa03aa12e3f00e045029082"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:VRNYWBAP4OGGSTHSNUUWSTB2UH","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Vero: An Open RL Recipe for General Visual Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Open reinforcement learning with broad visual data builds general reasoners that rival closed models.","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CV","authors_text":"Danqi Chen, Gabriel Sarch, Haoyang Wu, Linrong Cai, Qunzhong Wang, Zhuang Liu","submitted_at":"2026-04-06T17:56:25Z","abstract_excerpt":"What does it take to build a visual reasoner that works across charts, science, spatial understanding, and open-ended tasks? The strongest vision-language models (VLMs) suggest that broad visual reasoning is within reach, yet their closed data and reinforcement learning (RL) pipelines make their gains difficult to study, reproduce, or extend. We introduce Vero, a family of fully open VLMs that match or exceed existing open-weight models across diverse visual reasoning tasks. We scale RL data and rewards across six broad task categories, constructing Vero-600K, a 600K-sample dataset from 59 dat"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Vero achieves state-of-the-art performance, improving over four base models by 3.6-5.3 points on average across VeroEval, our suite of 30 challenging benchmarks. Starting from Qwen3-VL-8B-Instruct, Vero outperforms Qwen3-VL-8B-Thinking on 23 of 30 benchmarks without additional proprietary thinking data.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the observed gains are primarily driven by broad data coverage and task-routed rewards rather than other unstated factors such as training hyperparameters, base model choice, or evaluation contamination; the abstract provides no controls or ablations to isolate these effects.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Vero is an open VLM family trained via RL on Vero-600K (600K samples from 59 datasets across six categories) with task-routed rewards, achieving SOTA gains of 3.6-5.3 points on 30 visual reasoning benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Open reinforcement learning with broad visual data builds general reasoners that rival closed models.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"1da080e6493b5fec52de26df18ba5d888da1a250183a06edf062d3a0f7415184"},"source":{"id":"2604.04917","kind":"arxiv","version":3},"verdict":{"id":"9763b334-90c8-4df8-87bb-46dbbcfbb155","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T19:08:48.299108Z","strongest_claim":"Vero achieves state-of-the-art performance, improving over four base models by 3.6-5.3 points on average across VeroEval, our suite of 30 challenging benchmarks. Starting from Qwen3-VL-8B-Instruct, Vero outperforms Qwen3-VL-8B-Thinking on 23 of 30 benchmarks without additional proprietary thinking data.","one_line_summary":"Vero is an open VLM family trained via RL on Vero-600K (600K samples from 59 datasets across six categories) with task-routed rewards, achieving SOTA gains of 3.6-5.3 points on 30 visual reasoning benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the observed gains are primarily driven by broad data coverage and task-routed rewards rather than other unstated factors such as training hyperparameters, base model choice, or evaluation contamination; the abstract provides no controls or ablations to isolate these effects.","pith_extraction_headline":"Open reinforcement learning with broad visual data builds general reasoners that rival closed models."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.04917/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":"e310d4d64693b486077c92cd71560b1f00fea0dd98b54eb8c0aa4341eee2b5dd"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"9763b334-90c8-4df8-87bb-46dbbcfbb155"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-19T16:12:53Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Mkdj4gTiKSMgEHM+kcFSsl6Qs3pOPTXnlJhVRV666vfA60jIeGIHhoWYjhhizwob54+CrnEa+WJY/GVfprENBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-30T14:00:19.293345Z"},"content_sha256":"0144fcb16eb42369b3cfbfbb97922f5e2e07f597f6f3bd63bed5a1796966cfbd","schema_version":"1.0","event_id":"sha256:0144fcb16eb42369b3cfbfbb97922f5e2e07f597f6f3bd63bed5a1796966cfbd"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/VRNYWBAP4OGGSTHSNUUWSTB2UH/bundle.json","state_url":"https://pith.science/pith/VRNYWBAP4OGGSTHSNUUWSTB2UH/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/VRNYWBAP4OGGSTHSNUUWSTB2UH/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":"2026-06-30T14:00:19Z","links":{"resolver":"https://pith.science/pith/VRNYWBAP4OGGSTHSNUUWSTB2UH","bundle":"https://pith.science/pith/VRNYWBAP4OGGSTHSNUUWSTB2UH/bundle.json","state":"https://pith.science/pith/VRNYWBAP4OGGSTHSNUUWSTB2UH/state.json","well_known_bundle":"https://pith.science/.well-known/pith/VRNYWBAP4OGGSTHSNUUWSTB2UH/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:VRNYWBAP4OGGSTHSNUUWSTB2UH","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":"02f3aeeaf574c5c19b0719a3eda8ffe161688a6249952d286b9a3b739270eeeb","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-06T17:56:25Z","title_canon_sha256":"b239fcf64cf625c6e1202214cd9c444a5428698b2729a6af594559fb0672bc09"},"schema_version":"1.0","source":{"id":"2604.04917","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.04917","created_at":"2026-06-19T16:12:53Z"},{"alias_kind":"arxiv_version","alias_value":"2604.04917v3","created_at":"2026-06-19T16:12:53Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.04917","created_at":"2026-06-19T16:12:53Z"},{"alias_kind":"pith_short_12","alias_value":"VRNYWBAP4OGG","created_at":"2026-06-19T16:12:53Z"},{"alias_kind":"pith_short_16","alias_value":"VRNYWBAP4OGGSTHS","created_at":"2026-06-19T16:12:53Z"},{"alias_kind":"pith_short_8","alias_value":"VRNYWBAP","created_at":"2026-06-19T16:12:53Z"}],"graph_snapshots":[{"event_id":"sha256:0144fcb16eb42369b3cfbfbb97922f5e2e07f597f6f3bd63bed5a1796966cfbd","target":"graph","created_at":"2026-06-19T16:12:53Z","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":"Vero achieves state-of-the-art performance, improving over four base models by 3.6-5.3 points on average across VeroEval, our suite of 30 challenging benchmarks. Starting from Qwen3-VL-8B-Instruct, Vero outperforms Qwen3-VL-8B-Thinking on 23 of 30 benchmarks without additional proprietary thinking data."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the observed gains are primarily driven by broad data coverage and task-routed rewards rather than other unstated factors such as training hyperparameters, base model choice, or evaluation contamination; the abstract provides no controls or ablations to isolate these effects."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Vero is an open VLM family trained via RL on Vero-600K (600K samples from 59 datasets across six categories) with task-routed rewards, achieving SOTA gains of 3.6-5.3 points on 30 visual reasoning benchmarks."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Open reinforcement learning with broad visual data builds general reasoners that rival closed models."}],"snapshot_sha256":"1da080e6493b5fec52de26df18ba5d888da1a250183a06edf062d3a0f7415184"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"e310d4d64693b486077c92cd71560b1f00fea0dd98b54eb8c0aa4341eee2b5dd"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2604.04917/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"What does it take to build a visual reasoner that works across charts, science, spatial understanding, and open-ended tasks? The strongest vision-language models (VLMs) suggest that broad visual reasoning is within reach, yet their closed data and reinforcement learning (RL) pipelines make their gains difficult to study, reproduce, or extend. We introduce Vero, a family of fully open VLMs that match or exceed existing open-weight models across diverse visual reasoning tasks. We scale RL data and rewards across six broad task categories, constructing Vero-600K, a 600K-sample dataset from 59 dat","authors_text":"Danqi Chen, Gabriel Sarch, Haoyang Wu, Linrong Cai, Qunzhong Wang, Zhuang Liu","cross_cats":["cs.AI","cs.CL"],"headline":"Open reinforcement learning with broad visual data builds general reasoners that rival closed models.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-06T17:56:25Z","title":"Vero: An Open RL Recipe for General Visual Reasoning"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2604.04917","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-10T19:08:48.299108Z","id":"9763b334-90c8-4df8-87bb-46dbbcfbb155","model_set":{"reader":"grok-4.3"},"one_line_summary":"Vero is an open VLM family trained via RL on Vero-600K (600K samples from 59 datasets across six categories) with task-routed rewards, achieving SOTA gains of 3.6-5.3 points on 30 visual reasoning benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Open reinforcement learning with broad visual data builds general reasoners that rival closed models.","strongest_claim":"Vero achieves state-of-the-art performance, improving over four base models by 3.6-5.3 points on average across VeroEval, our suite of 30 challenging benchmarks. Starting from Qwen3-VL-8B-Instruct, Vero outperforms Qwen3-VL-8B-Thinking on 23 of 30 benchmarks without additional proprietary thinking data.","weakest_assumption":"That the observed gains are primarily driven by broad data coverage and task-routed rewards rather than other unstated factors such as training hyperparameters, base model choice, or evaluation contamination; the abstract provides no controls or ablations to isolate these effects."}},"verdict_id":"9763b334-90c8-4df8-87bb-46dbbcfbb155"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:f5e23062947a9c15645a5c388c353899136b8b211aa03aa12e3f00e045029082","target":"record","created_at":"2026-06-19T16:12:53Z","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":"02f3aeeaf574c5c19b0719a3eda8ffe161688a6249952d286b9a3b739270eeeb","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-06T17:56:25Z","title_canon_sha256":"b239fcf64cf625c6e1202214cd9c444a5428698b2729a6af594559fb0672bc09"},"schema_version":"1.0","source":{"id":"2604.04917","kind":"arxiv","version":3}},"canonical_sha256":"ac5b8b040fe38c694cf26d29694c3aa1c040180d3d5e4a74933df1554b7f149f","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ac5b8b040fe38c694cf26d29694c3aa1c040180d3d5e4a74933df1554b7f149f","first_computed_at":"2026-06-19T16:12:53.850107Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-19T16:12:53.850107Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"nOeawuoq3LFdme++jrefQimTIVOTUh2++RJrQEvHuqrYJui5Oexq+ADQv4KgMFgWucsLwd827/PH0VnhJVaXCg==","signature_status":"signed_v1","signed_at":"2026-06-19T16:12:53.850520Z","signed_message":"canonical_sha256_bytes"},"source_id":"2604.04917","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:f5e23062947a9c15645a5c388c353899136b8b211aa03aa12e3f00e045029082","sha256:0144fcb16eb42369b3cfbfbb97922f5e2e07f597f6f3bd63bed5a1796966cfbd"],"state_sha256":"114740e03982b527c285dd0f3825494a4b52c39d77e4bbd79a29d7f16ea4e171"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"x0SXTNH72nPn0Vmehz0tuz0MWJ97WcPITjCCyvbXx9p0jOZiAx+WpJc0mWJpiZ4r3kRnoX/GBL+bYONKeUpwDQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-30T14:00:19.295473Z","bundle_sha256":"0c6c2d63f3f59600fba4ca520c5d808ec1fa2d59a0bcf2ea35d86b62777a5b68"}}