{"id":"dba590b4bbf63ef7d3a32d8b478e280ce5ea90e1981e81150d614de99d429596","title":"Vohra and Xu ask when vision-language models need to zoom","argument":"In their paper, Siddharth Vohra and Min Xu argue that image handling before a vision-language model sees a picture can erase small details—and that zooming can recover much of what is lost. A car 20 pixels wide in an 8000 by 6000 drone photo reaches Claude about 6 pixels wide, they report, while OpenAI's default rejects the photo. Across tests of seven models from four vendors, the smallest text read half the time grew from 6 pixels or less on a small image to a predicted 10 to 13 pixels on a large one at default settings.\n\nThe authors compare splitting a photo into tiles with a two-step zoom, where the model first identifies the region it needs. Tiling raised accuracy for Claude and Gemini; for GPT-6.1 Sol, avoiding rejections raised it from 50.7% to 83.7%. The two-step approach matched or beat tiles on Gemini and came within 2.5 points on Claude, using about 40% of their image tokens. The findings make the practical question not just whether to zoom, but when and how.\n\nThe paper also highlights a risk beyond missed details: when detail is lost, Gemini gives a confident wrong answer more often than Claude, though both often say they cannot tell. The authors report that tiling only after Claude replies unsure comes close to always tiling while using fewer image tokens. They turn their findings into five rules for when an agent should zoom.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"This reading is based on the arXiv page's abstract, not the full paper or its underlying data. The abstract describes tests on seven models from four vendors; it does not establish how the results transfer to other models, images, or uses.","speaker":{"name":"Vohra, Siddharth, Xu, Min","roleAtSource":"Article author"},"source":{"title":"Know When to Zoom: How Small Is Too Small for Vision-Language Models?","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.22709","publishedAt":"2026-11-17","dateBasis":"Date basis not provided."},"evidence":[{"text":"Vohra and Xu report that image resizing or rejection can make small details harder for vision-language models to read.","excerpt":"Online vision-language models often miss small details in large images, because the service shrinks the image, or rejects it, before the model ever sees it.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"},{"text":"The authors tested seven models from four vendors using printed-code reading and questions about small objects in real photos.","excerpt":"We measure how much detail survives for seven models from four vendors, using an eye exam of printed codes and 601 questions about small objects in real photos.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"},{"text":"Tiling improved accuracy for Claude and Gemini, and avoiding rejections improved GPT-6.1 Sol's accuracy.","excerpt":"Splitting each photo into up to nine tiles raises accuracy by 11.1 to 14.5 percentage points for Claude and Gemini, and by avoiding rejections it lifts GPT-6.1 Sol from 50.7% to 83.7%.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"},{"text":"A two-step zoom matched or beat tiling on Gemini and came close on Claude while using fewer image tokens.","excerpt":"A two-step zoom, in which the model first names the region it needs, matches or beats tiles on Gemini and comes within 2.5 points of them on Claude, with about 40% of their image tokens.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"},{"text":"When detail is lost, the authors report that Gemini gives confident wrong answers more often than Claude.","excerpt":"Gemini gives a confident wrong answer in 43% to 48% of these cases, Claude in only 5% to 9%.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"},{"text":"The abstract reports that tiling only after an unsure Claude reply came close to always tiling while using fewer image tokens.","excerpt":"tiling only after an unsure reply comes within 1.4 points of always tiling with 62% to 70% of the image tokens.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22709","checkedAt":"2026-11-18T17:48:12.720Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well do the reported zoom strategies and model differences carry over to images and uses beyond the tests described in the abstract?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Vohra and Xu ask when vision-language models need to zoom","summary":"In their paper, Siddharth Vohra and Min Xu argue that image handling before a vision-language model sees a picture can erase small details—and that zooming can recover much of what is lost. A car 20 pixels wide in an 8000 by 6000 drone photo reaches Claude about 6 pixels wide, they report, while OpenAI's default rejects the photo. Across tests of seven models from four vendors, the smallest text read half the time grew from 6 pixels or less on a small image to a predicted 10 to 13 pixels on a large one at default settings.\n\nThe authors compare splitting a photo into tiles with a two-step zoom, where the model first identifies the region it needs. Tiling raised accuracy for Claude and Gemini; for GPT-6.1 Sol, avoiding rejections raised it from 50.7% to 83.7%. The two-step approach matched or beat tiles on Gemini and came within 2.5 points on Claude, using about 40% of their image tokens. The findings make the practical question not just whether to zoom, but when and how.\n\nThe paper also highlights a risk beyond missed details: when detail is lost, Gemini gives a confident wrong answer more often than Claude, though both often say they cannot tell. The authors report that tiling only after Claude replies unsure comes close to always tiling while using fewer image tokens. They turn their findings into five rules for when an agent should zoom.","question":"How well do the reported zoom strategies and model differences carry over to images and uses beyond the tests described in the abstract?","claims":[{"text":"Vohra and Xu report that image resizing or rejection can make small details harder for vision-language models to read.","excerpt":"Online vision-language models often miss small details in large images, because the service shrinks the image, or rejects it, before the model ever sees it.","support":"source-stated"},{"text":"The authors tested seven models from four vendors using printed-code reading and questions about small objects in real photos.","excerpt":"We measure how much detail survives for seven models from four vendors, using an eye exam of printed codes and 601 questions about small objects in real photos.","support":"source-stated"},{"text":"Tiling improved accuracy for Claude and Gemini, and avoiding rejections improved GPT-6.1 Sol's accuracy.","excerpt":"Splitting each photo into up to nine tiles raises accuracy by 11.1 to 14.5 percentage points for Claude and Gemini, and by avoiding rejections it lifts GPT-6.1 Sol from 50.7% to 83.7%.","support":"source-stated"},{"text":"A two-step zoom matched or beat tiling on Gemini and came close on Claude while using fewer image tokens.","excerpt":"A two-step zoom, in which the model first names the region it needs, matches or beats tiles on Gemini and comes within 2.5 points of them on Claude, with about 40% of their image tokens.","support":"source-stated"},{"text":"When detail is lost, the authors report that Gemini gives confident wrong answers more often than Claude.","excerpt":"Gemini gives a confident wrong answer in 43% to 48% of these cases, Claude in only 5% to 9%.","support":"source-stated"},{"text":"The abstract reports that tiling only after an unsure Claude reply came close to always tiling while using fewer image tokens.","excerpt":"tiling only after an unsure reply comes within 1.4 points of always tiling with 62% to 70% of the image tokens.","support":"source-stated"}],"limitations":["This reading is based on the arXiv page's abstract, not the full paper or its underlying data.","The abstract describes tests on seven models from four vendors; it does not establish how the results transfer to other models, images, or uses."],"mediaId":"b954310309bc9fd0b1224b494cc8a068517b316aa115a5f0976e0364055c4e64","correctionReason":null,"id":"dba590b4bbf63ef7d3a32d8b478e280ce5ea90e1981e81150d614de99d429596","publicationId":"7aa40093-41ff-42ec-8288-3ba48f9c902f","version":1,"sourceUrl":"https://arxiv.org/abs/2611.22709","sourceTitle":"Know When to Zoom: How Small Is Too Small for Vision-Language Models?","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-17","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-18T05:07:18.650Z","createdAt":"2026-11-18T17:16:44.070Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"b954310309bc9fd0b1224b494cc8a068517b316aa115a5f0976e0364055c4e64","kind":"article","url":"https://arxiv.org/abs/2611.22709","sourceUrl":"https://arxiv.org/abs/2611.22709","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-18T17:06:26.453Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22709","checkedAt":"2026-11-18T17:48:12.720Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v9","mediaId":"b954310309bc9fd0b1224b494cc8a068517b316aa115a5f0976e0364055c4e64","confidence":0.7,"provider":"typesafe","model":"jev-1.13.0","inputHash":"b3e57223877e7cda0f9b2eba8bd7a872105ffd6f344fbdf595aed5f2477cddc1","decidedAt":"2026-11-18T17:06:29.028Z"},"provenance":{"textHash":"b0c0e43eb0ccea1189fde0c9eff057817ccb611025a747438be8a0044634d768","observationId":"68f9b6d6dea070d5faaae6be0055979ec4d00b299786c122b2551240dfb796c6","retrievedAt":"2026-11-18T17:06:26.064Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"c9930cd4-536d-4aa6-82bd-6ecbff7121d5"},"reviewer":{"model":"gpt-6-luna","runId":"7f56bc5d-f806-4c97-8a64-a4875414cbf1"}},"audit":{"status":"passed","reviewedAt":"2026-11-18T17:16:44.070Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Vohra, Siddharth","url":null,"kind":"unknown"},{"name":"Xu, Min","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Vohra, Siddharth","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Vohra, Siddharth","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22709","locator":"head/meta[name=author|citation_author]"},"assertionId":"2adb90f815237576ac7426a1dfc095c08a38a82239a913122baf401b29ad2273:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T17:48:12.720Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"2adb90f815237576ac7426a1dfc095c08a38a82239a913122baf401b29ad2273","sourceUrl":"https://arxiv.org/abs/2611.22709","contentSha256":"833fd4442ed88d51b2ee7ae6aa679192af6960d52a23dea60562ffc815c4a7b3","textHash":"b0c0e43eb0ccea1189fde0c9eff057817ccb611025a747438be8a0044634d768","observedAt":"2026-11-18T17:48:12.720Z"}},{"actor":{"name":"Xu, Min","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Xu, Min","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22709","locator":"head/meta[name=author|citation_author]"},"assertionId":"2adb90f815237576ac7426a1dfc095c08a38a82239a913122baf401b29ad2273:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T17:48:12.720Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"2adb90f815237576ac7426a1dfc095c08a38a82239a913122baf401b29ad2273","sourceUrl":"https://arxiv.org/abs/2611.22709","contentSha256":"833fd4442ed88d51b2ee7ae6aa679192af6960d52a23dea60562ffc815c4a7b3","textHash":"b0c0e43eb0ccea1189fde0c9eff057817ccb611025a747438be8a0044634d768","observedAt":"2026-11-18T17:48:12.720Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.22709","snapshotId":"2adb90f815237576ac7426a1dfc095c08a38a82239a913122baf401b29ad2273","contentSha256":"833fd4442ed88d51b2ee7ae6aa679192af6960d52a23dea60562ffc815c4a7b3","textHash":"b0c0e43eb0ccea1189fde0c9eff057817ccb611025a747438be8a0044634d768","observedAt":"2026-11-18T17:48:12.720Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/dba590b4bbf63ef7d3a32d8b478e280ce5ea90e1981e81150d614de99d429596/source"}},"canonicalUrl":"https://escapemeridian.com/readings/vohra-and-xu-ask-when-vision-language-models-need-to-zoom-dba590b4","formats":{"html":"/readings/vohra-and-xu-ask-when-vision-language-models-need-to-zoom-dba590b4","markdown":"/readings/vohra-and-xu-ask-when-vision-language-models-need-to-zoom-dba590b4.md","json":"/readings/vohra-and-xu-ask-when-vision-language-models-need-to-zoom-dba590b4.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"conditions":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"},{"id":"claude","name":"Claude","kind":"model","context":"Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding.","referenceUrls":["https://claude.com/product/overview","https://www.anthropic.com/news","https://www.anthropic.com/research/team/economics"],"path":"/entities/claude","json":"/entities/claude.json","markdown":"/entities/claude.md"},{"id":"openai","name":"OpenAI","kind":"organization","context":"OpenAI develops AI models and products and publishes research, release announcements and evaluations.","referenceUrls":["https://openai.com/news/"],"path":"/entities/openai","json":"/entities/openai.json","markdown":"/entities/openai.md"},{"id":"time","name":"TIME","kind":"organization","context":"TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.","referenceUrls":["https://time.com/"],"path":"/entities/time","json":"/entities/time.json","markdown":"/entities/time.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"dba590b4bbf63ef7d3a32d8b478e280ce5ea90e1981e81150d614de99d429596","revision":"7aa40093-41ff-42ec-8288-3ba48f9c902f","revisionKind":"publication-id","title":"Vohra and Xu ask when vision-language models need to zoom","url":"https://escapemeridian.com/readings/vohra-and-xu-ask-when-vision-language-models-need-to-zoom-dba590b4"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.22709","revision":"b0c0e43eb0ccea1189fde0c9eff057817ccb611025a747438be8a0044634d768","revisionKind":"sha256-source-text","title":"Know When to Zoom: How Small Is Too Small for Vision-Language Models?","url":"https://arxiv.org/abs/2611.22709"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter."},"revisions":[{"id":"dba590b4bbf63ef7d3a32d8b478e280ce5ea90e1981e81150d614de99d429596","title":"Vohra and Xu ask when vision-language models need to zoom","argument":"In their paper, Siddharth Vohra and Min Xu argue that image handling before a vision-language model sees a picture can erase small details—and that zooming can recover much of what is lost. A car 20 pixels wide in an 8000 by 6000 drone photo reaches Claude about 6 pixels wide, they report, while OpenAI's default rejects the photo. Across tests of seven models from four vendors, the smallest text read half the time grew from 6 pixels or less on a small image to a predicted 10 to 13 pixels on a large one at default settings.\n\nThe authors compare splitting a photo into tiles with a two-step zoom, where the model first identifies the region it needs. Tiling raised accuracy for Claude and Gemini; for GPT-6.1 Sol, avoiding rejections raised it from 50.7% to 83.7%. The two-step approach matched or beat tiles on Gemini and came within 2.5 points on Claude, using about 40% of their image tokens. The findings make the practical question not just whether to zoom, but when and how.\n\nThe paper also highlights a risk beyond missed details: when detail is lost, Gemini gives a confident wrong answer more often than Claude, though both often say they cannot tell. The authors report that tiling only after Claude replies unsure comes close to always tiling while using fewer image tokens. They turn their findings into five rules for when an agent should zoom.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"This reading is based on the arXiv page's abstract, not the full paper or its underlying data. The abstract describes tests on seven models from four vendors; it does not establish how the results transfer to other models, images, or uses.","speaker":{"name":"Vohra, Siddharth, Xu, Min","roleAtSource":"Article author"},"source":{"title":"Know When to Zoom: How Small Is Too Small for Vision-Language Models?","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.22709","publishedAt":"2026-11-17","dateBasis":"Date basis not provided."},"evidence":[{"text":"Vohra and Xu report that image resizing or rejection can make small details harder for vision-language models to read.","excerpt":"Online vision-language models often miss small details in large images, because the service shrinks the image, or rejects it, before the model ever sees it.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"},{"text":"The authors tested seven models from four vendors using printed-code reading and questions about small objects in real photos.","excerpt":"We measure how much detail survives for seven models from four vendors, using an eye exam of printed codes and 601 questions about small objects in real photos.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"},{"text":"Tiling improved accuracy for Claude and Gemini, and avoiding rejections improved GPT-6.1 Sol's accuracy.","excerpt":"Splitting each photo into up to nine tiles raises accuracy by 11.1 to 14.5 percentage points for Claude and Gemini, and by avoiding rejections it lifts GPT-6.1 Sol from 50.7% to 83.7%.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"},{"text":"A two-step zoom matched or beat tiling on Gemini and came close on Claude while using fewer image tokens.","excerpt":"A two-step zoom, in which the model first names the region it needs, matches or beats tiles on Gemini and comes within 2.5 points of them on Claude, with about 40% of their image tokens.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"},{"text":"When detail is lost, the authors report that Gemini gives confident wrong answers more often than Claude.","excerpt":"Gemini gives a confident wrong answer in 43% to 48% of these cases, Claude in only 5% to 9%.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"},{"text":"The abstract reports that tiling only after an unsure Claude reply came close to always tiling while using fewer image tokens.","excerpt":"tiling only after an unsure reply comes within 1.4 points of always tiling with 62% to 70% of the image tokens.","url":"https://arxiv.org/abs/2611.22709","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22709","checkedAt":"2026-11-18T17:48:12.720Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well do the reported zoom strategies and model differences carry over to images and uses beyond the tests described in the abstract?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Vohra and Xu ask when vision-language models need to zoom","summary":"In their paper, Siddharth Vohra and Min Xu argue that image handling before a vision-language model sees a picture can erase small details—and that zooming can recover much of what is lost. A car 20 pixels wide in an 8000 by 6000 drone photo reaches Claude about 6 pixels wide, they report, while OpenAI's default rejects the photo. Across tests of seven models from four vendors, the smallest text read half the time grew from 6 pixels or less on a small image to a predicted 10 to 13 pixels on a large one at default settings.\n\nThe authors compare splitting a photo into tiles with a two-step zoom, where the model first identifies the region it needs. Tiling raised accuracy for Claude and Gemini; for GPT-6.1 Sol, avoiding rejections raised it from 50.7% to 83.7%. The two-step approach matched or beat tiles on Gemini and came within 2.5 points on Claude, using about 40% of their image tokens. The findings make the practical question not just whether to zoom, but when and how.\n\nThe paper also highlights a risk beyond missed details: when detail is lost, Gemini gives a confident wrong answer more often than Claude, though both often say they cannot tell. The authors report that tiling only after Claude replies unsure comes close to always tiling while using fewer image tokens. They turn their findings into five rules for when an agent should zoom.","question":"How well do the reported zoom strategies and model differences carry over to images and uses beyond the tests described in the abstract?","claims":[{"text":"Vohra and Xu report that image resizing or rejection can make small details harder for vision-language models to read.","excerpt":"Online vision-language models often miss small details in large images, because the service shrinks the image, or rejects it, before the model ever sees it.","support":"source-stated"},{"text":"The authors tested seven models from four vendors using printed-code reading and questions about small objects in real photos.","excerpt":"We measure how much detail survives for seven models from four vendors, using an eye exam of printed codes and 601 questions about small objects in real photos.","support":"source-stated"},{"text":"Tiling improved accuracy for Claude and Gemini, and avoiding rejections improved GPT-6.1 Sol's accuracy.","excerpt":"Splitting each photo into up to nine tiles raises accuracy by 11.1 to 14.5 percentage points for Claude and Gemini, and by avoiding rejections it lifts GPT-6.1 Sol from 50.7% to 83.7%.","support":"source-stated"},{"text":"A two-step zoom matched or beat tiling on Gemini and came close on Claude while using fewer image tokens.","excerpt":"A two-step zoom, in which the model first names the region it needs, matches or beats tiles on Gemini and comes within 2.5 points of them on Claude, with about 40% of their image tokens.","support":"source-stated"},{"text":"When detail is lost, the authors report that Gemini gives confident wrong answers more often than Claude.","excerpt":"Gemini gives a confident wrong answer in 43% to 48% of these cases, Claude in only 5% to 9%.","support":"source-stated"},{"text":"The abstract reports that tiling only after an unsure Claude reply came close to always tiling while using fewer image tokens.","excerpt":"tiling only after an unsure reply comes within 1.4 points of always tiling with 62% to 70% of the image tokens.","support":"source-stated"}],"limitations":["This reading is based on the arXiv page's abstract, not the full paper or its underlying data.","The abstract describes tests on seven models from four vendors; it does not establish how the results transfer to other models, images, or uses."],"mediaId":"b954310309bc9fd0b1224b494cc8a068517b316aa115a5f0976e0364055c4e64","correctionReason":null,"id":"dba590b4bbf63ef7d3a32d8b478e280ce5ea90e1981e81150d614de99d429596","publicationId":"7aa40093-41ff-42ec-8288-3ba48f9c902f","version":1,"sourceUrl":"https://arxiv.org/abs/2611.22709","sourceTitle":"Know When to Zoom: How Small Is Too Small for Vision-Language Models?","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-17","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-18T05:07:18.650Z","createdAt":"2026-11-18T17:16:44.070Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"b954310309bc9fd0b1224b494cc8a068517b316aa115a5f0976e0364055c4e64","kind":"article","url":"https://arxiv.org/abs/2611.22709","sourceUrl":"https://arxiv.org/abs/2611.22709","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-18T17:06:26.453Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.22709","checkedAt":"2026-11-18T17:48:12.720Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v9","mediaId":"b954310309bc9fd0b1224b494cc8a068517b316aa115a5f0976e0364055c4e64","confidence":0.7,"provider":"typesafe","model":"jev-1.13.0","inputHash":"b3e57223877e7cda0f9b2eba8bd7a872105ffd6f344fbdf595aed5f2477cddc1","decidedAt":"2026-11-18T17:06:29.028Z"},"provenance":{"textHash":"b0c0e43eb0ccea1189fde0c9eff057817ccb611025a747438be8a0044634d768","observationId":"68f9b6d6dea070d5faaae6be0055979ec4d00b299786c122b2551240dfb796c6","retrievedAt":"2026-11-18T17:06:26.064Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"c9930cd4-536d-4aa6-82bd-6ecbff7121d5"},"reviewer":{"model":"gpt-6-luna","runId":"7f56bc5d-f806-4c97-8a64-a4875414cbf1"}},"audit":{"status":"passed","reviewedAt":"2026-11-18T17:16:44.070Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Vohra, Siddharth","url":null,"kind":"unknown"},{"name":"Xu, Min","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Vohra, Siddharth","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Vohra, Siddharth","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22709","locator":"head/meta[name=author|citation_author]"},"assertionId":"2adb90f815237576ac7426a1dfc095c08a38a82239a913122baf401b29ad2273:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T17:48:12.720Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"2adb90f815237576ac7426a1dfc095c08a38a82239a913122baf401b29ad2273","sourceUrl":"https://arxiv.org/abs/2611.22709","contentSha256":"833fd4442ed88d51b2ee7ae6aa679192af6960d52a23dea60562ffc815c4a7b3","textHash":"b0c0e43eb0ccea1189fde0c9eff057817ccb611025a747438be8a0044634d768","observedAt":"2026-11-18T17:48:12.720Z"}},{"actor":{"name":"Xu, Min","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Xu, Min","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.22709","locator":"head/meta[name=author|citation_author]"},"assertionId":"2adb90f815237576ac7426a1dfc095c08a38a82239a913122baf401b29ad2273:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-18T17:48:12.720Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"2adb90f815237576ac7426a1dfc095c08a38a82239a913122baf401b29ad2273","sourceUrl":"https://arxiv.org/abs/2611.22709","contentSha256":"833fd4442ed88d51b2ee7ae6aa679192af6960d52a23dea60562ffc815c4a7b3","textHash":"b0c0e43eb0ccea1189fde0c9eff057817ccb611025a747438be8a0044634d768","observedAt":"2026-11-18T17:48:12.720Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.22709","snapshotId":"2adb90f815237576ac7426a1dfc095c08a38a82239a913122baf401b29ad2273","contentSha256":"833fd4442ed88d51b2ee7ae6aa679192af6960d52a23dea60562ffc815c4a7b3","textHash":"b0c0e43eb0ccea1189fde0c9eff057817ccb611025a747438be8a0044634d768","observedAt":"2026-11-18T17:48:12.720Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/dba590b4bbf63ef7d3a32d8b478e280ce5ea90e1981e81150d614de99d429596/source"}}}],"audits":[{"publicationId":"7aa40093-41ff-42ec-8288-3ba48f9c902f","verdict":"passed","findings":[{"dimension":"attribution","status":"pass","explanation":"The paper and its findings are correctly attributed to Siddharth Vohra and Min Xu, the listed authors. Results are framed as what the authors report or argue, not as established fact.","excerpt":"Authors:Siddharth Vohra, Min Xu"},{"dimension":"summary","status":"pass","explanation":"The reading leads with the actors and the core argument: services shrink or reject large images before the model sees them, and zooming recovers detail. It explains tiling and two-step zoom in plain terms, and keeps the key numbers, the comparison of failure modes and the five rules. Leaving out the finding that a size-based pre-call check misses most of the gain is acceptable. The title and question are neutral. Image tokens could use a tooltip, but the term does not block understanding.","excerpt":"Zooming wins much of the detail back."},{"dimension":"date","status":"pass","explanation":"evidence.originalPublished is 2026-11-17. This matches the arXiv v1 submission date, not the later feed date.","excerpt":"[Submitted on 17 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Every claim matches the abstract, including the figures: 50.7% to 83.7%, within 2.5 points, about 40% of tokens, and 62% to 70% of tokens. The Gemini vs Claude contrast in confident wrong answers is supported. The line that the practical question is when and how to zoom follows from the authors' five rules. The limitations correctly note that the reading rests on the abstract only.","excerpt":"Gemini gives a confident wrong answer in 43% to 48% of these cases, Claude in only 5% to 9%."},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the original source.","excerpt":null}],"createdAt":"2026-11-18T17:16:44.070Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}