{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"openrouter-baidu-ernie-4.5-vl-28b-a3b","slug":"baidu-ernie-4.5-vl-28b-a3b","name":"Baidu: ERNIE 4.5 VL 28B A3B","type":"model","url":"https://openrouter.ai/models/baidu~ernie-4.5-vl-28b-a3b","page_url":"https://unfragile.ai/baidu-ernie-4.5-vl-28b-a3b","categories":["image-generation"],"tags":["baidu","api-access","text","image"],"pricing":{"model":"paid","free":false,"starting_price":"$1.40e-7 per prompt token"},"status":"active","verified":false},"capabilities":[{"id":"openrouter-baidu-ernie-4.5-vl-28b-a3b__cap_0","uri":"capability://image.visual.multimodal.text.image.understanding.with.heterogeneous.moe.routing","name":"multimodal text-image understanding with heterogeneous moe routing","description":"Processes both text and image inputs simultaneously using a 28B parameter Mixture-of-Experts architecture where only 3B parameters activate per token. Implements modality-isolated routing, meaning separate expert pathways handle text and vision features before fusion, enabling specialized processing for each modality without forcing them through identical computational paths. This heterogeneous MoE design allows the model to maintain distinct reasoning chains for language and vision while sharing a unified token-level gating mechanism.","intents":["I need to analyze images with detailed text descriptions and ask follow-up questions about visual content","I want to perform visual question answering where the model understands both the image context and nuanced text queries","I need to extract structured information from documents that contain both text and visual elements","I want to compare multiple images and reason about their relationships using natural language"],"best_for":["teams building document intelligence systems requiring simultaneous text-image understanding","developers creating multimodal RAG pipelines that need efficient inference with lower latency","enterprises processing mixed-media content (PDFs, screenshots, diagrams) at scale"],"limitations":["Modality-isolated routing adds architectural complexity — debugging cross-modality failures requires understanding expert specialization patterns","3B activated parameters per token means reduced per-token capacity compared to dense 28B models; may struggle with extremely long reasoning chains requiring full model width","No information on maximum image resolution or batch processing capabilities — likely constrained by token limits","MoE routing overhead introduces variable latency depending on expert load balancing; throughput may degrade under concurrent requests"],"requires":["API access via OpenRouter or Baidu's platform with valid authentication credentials","Images in standard formats (JPEG, PNG, WebP, GIF) with reasonable resolution (typically <4K recommended for inference efficiency)","Text prompts formatted for multimodal context (image descriptions or visual reasoning queries)"],"input_types":["text (natural language queries, prompts, instructions)","image (JPEG, PNG, WebP, GIF, potentially PDF pages as images)"],"output_types":["text (natural language responses, descriptions, answers)","structured text (JSON, markdown formatted analysis)"],"categories":["image-visual","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-baidu-ernie-4.5-vl-28b-a3b__cap_1","uri":"capability://image.visual.visual.question.answering.with.contextual.image.reasoning","name":"visual question answering with contextual image reasoning","description":"Answers natural language questions about image content by grounding language understanding in visual features extracted through the vision expert pathway. The model performs token-level fusion of image embeddings and text tokens, allowing it to generate answers that reference specific visual regions or objects mentioned in questions. This capability leverages the modality-isolated routing to maintain separate visual reasoning before integrating with language generation.","intents":["I want to ask detailed questions about what's in an image and get accurate, contextual answers","I need to identify objects, text, or relationships within images using natural language queries","I want to verify claims about image content or ask comparative questions across visual elements"],"best_for":["developers building accessibility tools that describe images for visually impaired users","teams creating content moderation systems that need to understand image context and user queries","e-commerce platforms requiring product image analysis and customer question answering"],"limitations":["No explicit support for video input — only static images; temporal reasoning across frames not supported","Accuracy on fine-grained visual details (small text in images, precise measurements) depends on image resolution and may degrade with low-quality inputs","Context window limitations mean very long question-answer histories may lose earlier visual references"],"requires":["Image file in supported format (JPEG, PNG, WebP, GIF)","Natural language question or prompt in English or supported languages","API endpoint access with proper authentication"],"input_types":["image (visual content to analyze)","text (natural language questions or prompts)"],"output_types":["text (natural language answers with visual grounding)"],"categories":["image-visual","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-baidu-ernie-4.5-vl-28b-a3b__cap_2","uri":"capability://image.visual.document.image.analysis.with.text.vision.fusion","name":"document image analysis with text-vision fusion","description":"Analyzes documents, forms, and screenshots by simultaneously processing visual layout and text content through separate expert pathways that fuse at the token level. The model can extract structured information from documents (tables, forms, receipts) by understanding both the spatial arrangement of elements (vision pathway) and semantic meaning of text (text pathway). The heterogeneous MoE architecture allows it to specialize in document structure recognition without diluting text understanding capacity.","intents":["I need to extract data from scanned documents, invoices, or forms while preserving layout understanding","I want to convert document images into structured data (JSON, CSV) with high accuracy","I need to understand document hierarchy and relationships between text elements in complex layouts"],"best_for":["teams building document processing pipelines for financial, legal, or administrative documents","enterprises automating form processing and data extraction from paper or digital documents","developers creating document understanding APIs that need to handle mixed-quality scans"],"limitations":["Performance on handwritten text or non-standard fonts may be lower than on printed documents","Multi-page document processing requires sequential image submission; no native support for PDF batch processing","Structured output format (JSON, CSV) requires explicit prompting — no built-in schema validation or guaranteed format compliance","OCR accuracy depends on image quality; low-resolution or heavily skewed documents may produce incomplete extraction"],"requires":["Document image in JPEG, PNG, or WebP format with reasonable resolution (300+ DPI recommended)","Clear prompting about desired output format and structure","API access with authentication credentials"],"input_types":["image (document, form, receipt, or screenshot)","text (extraction instructions, desired output format specification)"],"output_types":["text (extracted text content)","structured data (JSON, CSV, markdown table format)"],"categories":["image-visual","data-processing-analysis"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-baidu-ernie-4.5-vl-28b-a3b__cap_3","uri":"capability://image.visual.image.captioning.and.description.generation","name":"image captioning and description generation","description":"Generates natural language descriptions and captions for images by processing visual features through the vision expert pathway and generating coherent text through the text expert pathway with token-level fusion. The model can produce captions at varying levels of detail (short captions, detailed descriptions, technical analysis) based on prompt instructions. The sparse activation pattern (3B/28B) allows efficient batch processing of image captioning tasks.","intents":["I want to automatically generate alt-text for images in web applications or documents","I need to create detailed descriptions of images for accessibility or content management purposes","I want to generate captions for social media or image galleries at scale"],"best_for":["content management systems requiring automated alt-text generation for accessibility compliance","social media platforms or image galleries needing bulk caption generation","accessibility-focused teams building tools for visually impaired users"],"limitations":["Generated captions may contain hallucinations or inaccuracies, especially for ambiguous or complex images","No explicit control over caption length or style beyond prompt engineering; no built-in templates or structured caption formats","Bias in training data may result in stereotypical or incomplete descriptions for certain image types","Batch processing throughput limited by token budget and concurrent request handling on API"],"requires":["Image in JPEG, PNG, WebP, or GIF format","Optional: prompt specifying caption style, length, or focus areas","API access with authentication"],"input_types":["image (visual content to caption)","text (optional style or length instructions)"],"output_types":["text (natural language caption or description)"],"categories":["image-visual","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-baidu-ernie-4.5-vl-28b-a3b__cap_4","uri":"capability://image.visual.conversational.multimodal.chat.with.image.context.persistence","name":"conversational multimodal chat with image context persistence","description":"Maintains multi-turn conversations where users can reference previously shared images and ask follow-up questions that build on earlier visual context. The model preserves image embeddings and visual understanding across conversation turns, allowing users to ask 'what was in that image from earlier?' or refine questions about previously analyzed images. The heterogeneous MoE routing maintains separate visual and text reasoning chains that can be reused across turns without reprocessing images.","intents":["I want to have a back-and-forth conversation about an image, asking clarifying questions and requesting different analyses","I need to compare multiple images across conversation turns and discuss relationships between them","I want to iteratively refine my understanding of image content through natural dialogue"],"best_for":["developers building interactive image analysis chatbots or assistants","teams creating customer support systems that handle image-based inquiries with multi-turn dialogue","research tools requiring iterative visual analysis and discussion"],"limitations":["Context window constraints limit the number of previous conversation turns and images that can be referenced simultaneously","No explicit mechanism for managing image cache or optimizing re-reference of earlier images — each turn may require re-encoding visual features","Conversation history grows with each turn, potentially causing latency increase in later turns due to longer context processing","No built-in memory persistence across sessions — conversation state must be managed externally"],"requires":["API access with session management capability","Initial image in supported format (JPEG, PNG, WebP, GIF)","Natural language prompts for each conversation turn","Context window sufficient for multi-turn history (typically 4K-8K tokens recommended)"],"input_types":["image (initial or new images to analyze)","text (natural language questions, follow-ups, refinements)"],"output_types":["text (conversational responses with visual grounding)"],"categories":["image-visual","text-generation-language","memory-knowledge"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-baidu-ernie-4.5-vl-28b-a3b__cap_5","uri":"capability://image.visual.cross.modal.semantic.understanding.and.reasoning","name":"cross-modal semantic understanding and reasoning","description":"Performs reasoning tasks that require simultaneous understanding of both text and visual semantics, such as determining if an image matches a text description, identifying contradictions between image content and text claims, or reasoning about abstract relationships between visual and textual information. The modality-isolated expert routing allows the model to develop independent semantic representations in each modality before fusion, enabling more nuanced cross-modal reasoning than models that force both modalities through identical pathways.","intents":["I need to verify if an image matches a product description or claim","I want to detect contradictions or inconsistencies between image content and accompanying text","I need to perform semantic matching between images and text queries for retrieval or ranking tasks"],"best_for":["content moderation teams detecting misleading image-text combinations or misinformation","e-commerce platforms validating product images against descriptions","search and retrieval systems requiring cross-modal semantic matching"],"limitations":["Reasoning accuracy depends on clarity of both visual and textual inputs; ambiguous images or vague descriptions reduce reliability","No explicit confidence scoring or uncertainty quantification — model outputs binary or categorical judgments without confidence metrics","Cross-modal hallucinations possible where model invents connections between image and text that don't actually exist","Limited to reasoning about explicit visual and textual content; implicit or cultural context may be missed"],"requires":["Image in supported format (JPEG, PNG, WebP, GIF)","Text description or claim to compare against image","Clear prompting about the reasoning task (matching, contradiction detection, etc.)","API access with authentication"],"input_types":["image (visual content)","text (description, claim, or query to reason about)"],"output_types":["text (reasoning explanation, match/mismatch judgment, semantic analysis)"],"categories":["image-visual","text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-baidu-ernie-4.5-vl-28b-a3b__cap_6","uri":"capability://image.visual.efficient.batch.processing.of.multimodal.requests","name":"efficient batch processing of multimodal requests","description":"Processes multiple image-text pairs or sequential multimodal requests efficiently through sparse MoE activation, where only 3B of 28B parameters activate per token. This enables higher throughput and lower latency for batch operations compared to dense models, making it suitable for processing large volumes of images with associated queries. The sparse activation pattern reduces memory footprint and computational cost per request, allowing more concurrent requests on the same hardware.","intents":["I need to process thousands of images with associated queries in a batch job","I want to minimize API costs for high-volume multimodal inference","I need to achieve low-latency responses for real-time multimodal applications at scale"],"best_for":["teams running batch processing jobs for document analysis, image captioning, or content moderation","cost-sensitive applications requiring high-volume multimodal inference","real-time systems needing low-latency multimodal responses (image search, visual QA APIs)"],"limitations":["Sparse activation may cause variable latency depending on expert load balancing and routing decisions","Batch processing throughput depends on API rate limits and concurrent request handling — no guaranteed SLA for batch operations","Memory efficiency gains from sparse activation may not translate to proportional cost savings if API pricing doesn't account for parameter activation","No built-in batch API endpoint — requires sequential or parallel requests through standard API"],"requires":["Multiple images in supported formats (JPEG, PNG, WebP, GIF)","Associated text queries or prompts for each image","API access with sufficient rate limit quota","Batch orchestration logic (external to model) for managing request queuing and result aggregation"],"input_types":["image (multiple images for batch processing)","text (queries or prompts for each image)"],"output_types":["text (responses for each image-query pair)"],"categories":["image-visual","text-generation-language","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":24,"verified":false,"data_access_risk":"low","permissions":["API access via OpenRouter or Baidu's platform with valid authentication credentials","Images in standard formats (JPEG, PNG, WebP, GIF) with reasonable resolution (typically <4K recommended for inference efficiency)","Text prompts formatted for multimodal context (image descriptions or visual reasoning queries)","Image file in supported format (JPEG, PNG, WebP, GIF)","Natural language question or prompt in English or supported languages","API endpoint access with proper authentication","Document image in JPEG, PNG, or WebP format with reasonable resolution (300+ DPI recommended)","Clear prompting about desired output format and structure","API access with authentication credentials","Image in JPEG, PNG, WebP, or GIF format"],"failure_modes":["Modality-isolated routing adds architectural complexity — debugging cross-modality failures requires understanding expert specialization patterns","3B activated parameters per token means reduced per-token capacity compared to dense 28B models; may struggle with extremely long reasoning chains requiring full model width","No information on maximum image resolution or batch processing capabilities — likely constrained by token limits","MoE routing overhead introduces variable latency depending on expert load balancing; throughput may degrade under concurrent requests","No explicit support for video input — only static images; temporal reasoning across frames not supported","Accuracy on fine-grained visual details (small text in images, precise measurements) depends on image resolution and may degrade with low-quality inputs","Context window limitations mean very long question-answer histories may lose earlier visual references","Performance on handwritten text or non-standard fonts may be lower than on printed documents","Multi-page document processing requires sequential image submission; no native support for PDF batch processing","Structured output format (JSON, CSV) requires explicit prompting — no built-in schema validation or guaranteed format compliance","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.05,"quality":0.39,"ecosystem":0.27,"match_graph":0.25,"freshness":0.75,"weights":{"adoption":0.35,"quality":0.2,"ecosystem":0.1,"match_graph":0.3,"freshness":0.05}},"observed_outcomes":{"matches":0,"success_rate":0,"avg_confidence":0,"top_intents":[],"last_matched_at":null},"maintenance":{"status":"active","updated_at":"2026-05-24T12:16:24.484Z","last_scraped_at":"2026-05-03T15:20:45.776Z","last_commit":null},"community":{"stars":null,"forks":null,"weekly_downloads":null,"model_downloads":null,"model_likes":null}},"distribution":{"claim_url":"https://unfragile.ai/submit?claim=baidu-ernie-4.5-vl-28b-a3b","compare_url":"https://unfragile.ai/compare?artifact=baidu-ernie-4.5-vl-28b-a3b"}},"signature":"yMuRLWHSa8zqer9dX0LvuEbUN1XQgBrYxPCisbUu7c8UjkrRuenMyj0gxrT1uF9kCnW9JyI8Mbw3hyqnZ6kvCg==","signedAt":"2026-06-19T21:12:56.724Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/baidu-ernie-4.5-vl-28b-a3b","artifact":"https://unfragile.ai/baidu-ernie-4.5-vl-28b-a3b","verify":"https://unfragile.ai/api/v1/verify?slug=baidu-ernie-4.5-vl-28b-a3b","publicKey":"https://unfragile.ai/api/v1/trust-passport-public-key","spec":"https://unfragile.ai/trust","schema":"https://unfragile.ai/schema.json","docs":"https://unfragile.ai/docs"}}