{"paper":{"title":"ATLAS: Agentic or Latent Visual Reasoning? One Word is Enough for Both","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"One discrete functional token suffices for both agentic operations and latent visual reasoning.","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CV","authors_text":"Pheng-Ann Heng, Rain Liu, Xinyan Chen, Ziyu Guo","submitted_at":"2026-05-14T17:59:55Z","abstract_excerpt":"Visual reasoning, often interleaved with intermediate visual states, has emerged as a promising direction in the field. A straightforward approach is to directly generate images via unified models during reasoning, but this is computationally expensive and architecturally non-trivial. Recent alternatives include agentic reasoning through code or tool calls, and latent reasoning with learnable hidden embeddings. However, agentic methods incur context-switching latency from external execution, while latent methods lack task generalization and are difficult to train with autoregressive paralleliz"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"ATLAS achieves superior performance on challenging benchmarks while maintaining clear interpretability by using a single discrete functional token that serves as both an agentic operation and a latent visual reasoning unit.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a single discrete functional token can effectively internalize visual operations without any visual supervision and still generalize across tasks when generated via standard next-token prediction.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ATLAS uses a single functional token to unify agentic and latent visual reasoning without image generation or external execution.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"One discrete functional token suffices for both agentic operations and latent visual reasoning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"54db4534ab32f0ea5bf9dc95c6c87a3c4b3e69a860ce4d6570c579b438f4a032"},"source":{"id":"2605.15198","kind":"arxiv","version":1},"verdict":{"id":"9f14be86-a865-4e9e-9c3a-131216c39a5d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T03:05:18.872222Z","strongest_claim":"ATLAS achieves superior performance on challenging benchmarks while maintaining clear interpretability by using a single discrete functional token that serves as both an agentic operation and a latent visual reasoning unit.","one_line_summary":"ATLAS uses a single functional token to unify agentic and latent visual reasoning without image generation or external execution.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a single discrete functional token can effectively internalize visual operations without any visual supervision and still generalize across tasks when generated via standard next-token prediction.","pith_extraction_headline":"One discrete functional token suffices for both agentic operations and latent visual reasoning."},"references":{"count":24,"sample":[{"doi":"","year":null,"title":"LLaVA-OneVision-1.5: Fully Open Framework for Democratized Multimodal Training","work_id":"41c2802e-aff9-482f-b506-10955ff0838d","ref_index":1,"cited_arxiv_id":"2509.23661","is_internal_anchor":true},{"doi":"","year":null,"title":"Qwen3-VL Technical Report","work_id":"1fe243aa-e3c0-4da6-b391-4cbcfc88d5c0","ref_index":2,"cited_arxiv_id":"2511.21631","is_internal_anchor":true},{"doi":"","year":null,"title":"Anole: An open, autoregressive, native large multimodal models for interleaved image-text generation","work_id":"31ce9d99-2071-41a0-9f51-51b8c5e3ba7e","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities","work_id":"008df105-2fdd-45d8-857a-8e35868aecb6","ref_index":4,"cited_arxiv_id":"2507.06261","is_internal_anchor":true},{"doi":"","year":null,"title":"Emerging Properties in Unified Multimodal Pretraining","work_id":"e0cfd82c-f5d4-44fd-b531-ec73ab0a805b","ref_index":5,"cited_arxiv_id":"2505.14683","is_internal_anchor":true}],"resolved_work":24,"snapshot_sha256":"59531425264020a4256c48ab840b47299f63cfc2a769c710677630731a0adee2","internal_anchors":18},"formal_canon":{"evidence_count":2,"snapshot_sha256":"affdd43ea14ce12d209cda4b098642c2992f038d3c83e22bfc1f7f6cec067d5e"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}