{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"openrouter-anthropic-claude-3-haiku","slug":"anthropic-claude-3-haiku","name":"Anthropic: Claude 3 Haiku","type":"model","url":"https://openrouter.ai/models/anthropic~claude-3-haiku","page_url":"https://unfragile.ai/anthropic-claude-3-haiku","categories":["model-training","testing-quality"],"tags":["anthropic","api-access","text","image"],"pricing":{"model":"paid","free":false,"starting_price":"$2.50e-7 per prompt token"},"status":"active","verified":false},"capabilities":[{"id":"openrouter-anthropic-claude-3-haiku__cap_0","uri":"capability://image.visual.multimodal.text.and.image.understanding.with.vision.encoding","name":"multimodal text and image understanding with vision encoding","description":"Claude 3 Haiku processes both text and image inputs through a unified transformer architecture with integrated vision encoding, enabling simultaneous analysis of visual and textual content. The model uses a shared token space where image patches are encoded into the same embedding dimension as text tokens, allowing cross-modal attention patterns to emerge naturally. This architecture enables the model to reason about relationships between visual elements and textual descriptions without separate modality-specific processing pipelines.","intents":["I need to analyze screenshots, diagrams, or photos alongside text queries in a single request","I want to extract structured data from documents that contain both text and images","I need to verify visual content matches textual descriptions or claims","I want to generate descriptions or summaries of images with contextual text understanding"],"best_for":["developers building document processing pipelines that handle mixed-media content","teams automating visual QA or content moderation workflows","builders creating accessibility tools that need to understand images in context"],"limitations":["Image resolution is limited to ~1568x1568 pixels; larger images are downsampled, potentially losing fine detail","No video frame extraction — must provide individual image frames as separate inputs","Image understanding latency adds ~100-200ms vs text-only inference due to vision encoding overhead","Cannot generate, edit, or manipulate images — vision is read-only"],"requires":["API key from Anthropic or OpenRouter","Images in JPEG, PNG, GIF, or WebP format","Base64 encoding for image transmission via API","HTTP/2 capable client for efficient multipart request handling"],"input_types":["text (UTF-8, up to context window limit)","image (JPEG, PNG, GIF, WebP, max ~1568x1568px)"],"output_types":["text (natural language response)","structured JSON (with appropriate prompting)","code snippets (when analyzing code in images)"],"categories":["image-visual","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-anthropic-claude-3-haiku__cap_1","uri":"capability://text.generation.language.fast.inference.with.optimized.model.compression.and.quantization","name":"fast inference with optimized model compression and quantization","description":"Claude 3 Haiku achieves sub-second response latency through architectural optimizations including knowledge distillation from larger Claude models, parameter-efficient fine-tuning, and inference-time optimizations like token batching and KV-cache management. The model uses a smaller parameter count than Claude 3 Sonnet while maintaining competitive accuracy through selective knowledge transfer and careful pruning of less-critical attention heads. Anthropic's inference infrastructure uses speculative decoding and dynamic batching to maximize throughput without sacrificing latency.","intents":["I need to build real-time chat applications where sub-second response time is critical","I want to process high-volume API requests with minimal per-token latency cost","I need to deploy an LLM on resource-constrained infrastructure or edge devices","I want to reduce inference costs while maintaining reasonable output quality for high-volume workloads"],"best_for":["startups and indie developers optimizing for cost-per-inference in high-volume scenarios","teams building real-time customer support chatbots or interactive applications","builders creating mobile or edge-deployed LLM applications with strict latency budgets","organizations processing millions of short-form requests (classification, tagging, extraction)"],"limitations":["Smaller effective context window (200K tokens) compared to Claude 3 Sonnet (200K) — no advantage here, but reasoning depth is shallower","Lower accuracy on complex multi-step reasoning tasks; performance degrades on problems requiring >5 reasoning steps","Cannot reliably handle very long documents (>50K tokens) with high accuracy due to attention pattern limitations","Knowledge cutoff date limits real-time information access; requires external retrieval for current events"],"requires":["API key from Anthropic or OpenRouter","Network connectivity for API calls (no local inference without custom deployment)","Minimum 1MB/s bandwidth for optimal streaming performance","HTTP/1.1 or HTTP/2 capable client library"],"input_types":["text (UTF-8, up to 200K tokens)","image (when using multimodal capability)"],"output_types":["text (streaming or non-streaming)","structured JSON (with schema constraints)","tool/function calls (via function-calling API)"],"categories":["text-generation-language","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-anthropic-claude-3-haiku__cap_10","uri":"capability://text.generation.language.few.shot.learning.with.in.context.examples.for.task.adaptation","name":"few-shot learning with in-context examples for task adaptation","description":"Claude 3 Haiku can adapt to new tasks by providing examples in the prompt (few-shot learning), without requiring fine-tuning or retraining. The model learns patterns from 1-10 examples and applies them to new inputs, enabling rapid task customization. This is implemented through the model's general language understanding — it recognizes the pattern in examples and generalizes to unseen inputs. Few-shot learning works across diverse tasks including classification, extraction, summarization, and code generation.","intents":["I need to customize the model's behavior for domain-specific tasks without fine-tuning","I want to teach the model a new classification scheme or output format with just a few examples","I need to adapt the model to project-specific terminology or conventions","I want to implement task-specific behavior that changes between requests"],"best_for":["developers building flexible applications that adapt to different use cases","teams implementing domain-specific NLP tasks without fine-tuning infrastructure","builders creating customizable APIs where clients can define their own task patterns","organizations experimenting with new tasks before committing to fine-tuning"],"limitations":["Few-shot learning performance plateaus around 5-10 examples; adding more examples doesn't significantly improve accuracy and wastes context tokens","Complex tasks requiring deep domain knowledge may not be learnable from examples alone; fine-tuning is more effective for specialized domains","Example quality significantly impacts performance; poorly chosen or ambiguous examples degrade accuracy","No explicit confidence scoring; the model may apply learned patterns incorrectly without indicating uncertainty"],"requires":["API key from Anthropic or OpenRouter","Well-chosen examples that clearly demonstrate the desired pattern","Clear task description or system prompt explaining the pattern","Validation data to assess whether few-shot learning is sufficient for the task"],"input_types":["text (task description, examples, and new inputs)"],"output_types":["text (output following the pattern demonstrated in examples)","structured data (if examples show structured output format)"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-anthropic-claude-3-haiku__cap_2","uri":"capability://text.generation.language.instruction.following.with.constitutional.ai.alignment","name":"instruction-following with constitutional ai alignment","description":"Claude 3 Haiku is trained using Constitutional AI (CAI), a technique where the model learns to follow a set of explicit principles (constitution) through self-critique and reinforcement learning. During inference, the model applies these learned principles to interpret user instructions accurately while refusing harmful requests, maintaining context-appropriate tone, and correcting its own errors when prompted. The alignment is baked into the model weights rather than applied as a post-hoc filter, enabling nuanced judgment about edge cases without rigid rule-based blocking.","intents":["I need a model that refuses harmful requests but still handles legitimate edge cases (e.g., discussing security vulnerabilities in educational context)","I want consistent, predictable behavior across diverse instruction types without excessive false-positive refusals","I need to build applications where the model can self-correct when given feedback about its mistakes","I want a model that adapts tone and formality based on context without explicit system prompts for each variation"],"best_for":["teams building customer-facing applications requiring nuanced safety without over-blocking","developers creating educational or research tools that discuss sensitive topics responsibly","builders implementing multi-turn conversations where the model needs to learn from user feedback","organizations in regulated industries (healthcare, finance) needing interpretable safety decisions"],"limitations":["Constitutional AI alignment is probabilistic — edge cases near refusal boundaries may produce inconsistent results across identical inputs due to sampling temperature","No explicit audit trail of which constitutional principle triggered a refusal; debugging safety decisions requires prompt engineering","Alignment is optimized for English; non-English languages may show weaker adherence to constitutional principles","Cannot be easily fine-tuned to adopt different constitutional principles without retraining; alignment is fixed at model release"],"requires":["API key from Anthropic or OpenRouter","Understanding of Constitutional AI principles to effectively prompt the model","Acceptance that some requests will be refused; no override mechanism for safety decisions","Monitoring and logging infrastructure to track refusal patterns in production"],"input_types":["text instructions (natural language, code, structured prompts)"],"output_types":["text (refusal explanation, corrected response, or compliant output)","structured data (when instruction specifies format)"],"categories":["text-generation-language","safety-moderation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-anthropic-claude-3-haiku__cap_3","uri":"capability://tool.use.integration.function.calling.with.schema.based.tool.binding","name":"function calling with schema-based tool binding","description":"Claude 3 Haiku supports structured function calling where developers define tools as JSON schemas, and the model learns to emit properly-formatted function calls within its text output. The model receives tool definitions at inference time (not training time), enabling dynamic tool composition without model retraining. The implementation uses a special token sequence to delimit function calls, allowing the model to interleave natural language responses with structured tool invocations in a single generation pass.","intents":["I want to build an agent that can call external APIs (database queries, REST endpoints, calculations) based on user requests","I need to extract structured data from unstructured text and validate it against a schema before processing","I want to create a multi-step workflow where the model decides which tools to call and in what order","I need to integrate Claude with my existing tool ecosystem without building custom parsing logic"],"best_for":["developers building LLM agents that orchestrate multiple APIs or microservices","teams creating data extraction pipelines that need schema validation","builders implementing autonomous workflows where the model decides tool sequencing","organizations integrating Claude into existing tool ecosystems (Zapier, Make, custom platforms)"],"limitations":["Function calls are generated as text tokens, not native binary format — parsing errors can occur if the model produces malformed JSON, requiring fallback handling","No built-in retry logic if a function call fails; the application must implement error handling and re-prompt the model","Tool definitions are sent with every request, adding ~500 bytes to ~2KB overhead per tool; large tool registries (>50 tools) may impact latency","No native support for streaming function calls — the entire call must be generated before execution, limiting real-time responsiveness"],"requires":["API key from Anthropic or OpenRouter","JSON schema definitions for each tool (following OpenAI function-calling format or Anthropic's tool_use format)","Application-level error handling for malformed function calls","HTTP client capable of making calls to external APIs after Claude generates function calls"],"input_types":["text (user request)","JSON schema (tool definitions)"],"output_types":["text (natural language response)","function calls (JSON-formatted tool invocations)","mixed (interleaved text and function calls)"],"categories":["tool-use-integration","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-anthropic-claude-3-haiku__cap_4","uri":"capability://memory.knowledge.context.window.management.with.200k.token.capacity","name":"context window management with 200k token capacity","description":"Claude 3 Haiku supports a 200,000 token context window, enabling the model to process entire documents, codebases, or conversation histories in a single request without chunking or summarization. The implementation uses efficient attention mechanisms (likely including sparse attention or sliding window patterns) to manage the computational cost of long contexts. Tokens are counted consistently across text and images, with images typically consuming 100-300 tokens depending on resolution and complexity.","intents":["I need to analyze entire documents (research papers, legal contracts, codebases) without splitting them into chunks","I want to maintain full conversation history for context-aware multi-turn interactions without summarization","I need to process multiple files together and reason about relationships between them","I want to provide comprehensive system prompts, examples, and reference materials without consuming user token budgets"],"best_for":["developers building document analysis tools (legal review, code auditing, research synthesis)","teams creating long-form conversational agents with persistent memory","builders implementing RAG systems where full document context improves accuracy","organizations processing large codebases for refactoring, migration, or analysis"],"limitations":["Latency increases non-linearly with context length; 200K token requests may take 5-10x longer than 10K token requests","Cost scales linearly with input tokens; a 200K token request costs ~20x more than a 10K token request","Model accuracy degrades on tasks requiring reasoning about information in the middle of very long contexts (the 'lost in the middle' problem affects even long-context models)","No built-in context compression or summarization; developers must manage context manually to avoid exceeding limits"],"requires":["API key from Anthropic or OpenRouter","Sufficient API quota/credits to handle large token counts (200K tokens at standard pricing may cost $0.50-$2.00 per request)","Client library that supports streaming to handle long response times gracefully","Tokenizer compatible with Claude's token counting (Anthropic provides reference implementations)"],"input_types":["text (up to 200K tokens)","image (multiple images, each consuming 100-300 tokens)"],"output_types":["text (up to 4K tokens output limit per request)","structured data (JSON, code, etc.)"],"categories":["memory-knowledge","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-anthropic-claude-3-haiku__cap_5","uri":"capability://text.generation.language.streaming.response.generation.with.token.by.token.output","name":"streaming response generation with token-by-token output","description":"Claude 3 Haiku supports streaming inference where tokens are emitted one at a time as they are generated, enabling real-time display of responses to users before generation completes. The streaming implementation uses Server-Sent Events (SSE) over HTTP, with each token wrapped in a JSON event. This allows applications to display partial responses immediately, improving perceived latency and enabling cancellation of long-running generations.","intents":["I want to display model responses in real-time as they're generated, improving user experience in chat applications","I need to cancel long-running generations if the user stops waiting or requests a different query","I want to process tokens as they arrive for downstream processing (e.g., text-to-speech synthesis of partial responses)","I need to implement progressive disclosure of information in interactive applications"],"best_for":["developers building real-time chat interfaces or conversational UIs","teams creating interactive applications where perceived latency matters more than actual latency","builders implementing text-to-speech or other streaming-dependent features","organizations with bandwidth constraints that benefit from progressive token delivery"],"limitations":["Streaming adds ~50-100ms overhead compared to non-streaming due to HTTP framing and event serialization","Function calls cannot be streamed — the entire call must be generated before it's available, limiting real-time tool orchestration","Token-level streaming makes it difficult to implement token-level retry logic; errors in the middle of a stream cannot be recovered","Client-side buffering is required to reconstruct the full response; naive implementations may lose tokens if the connection drops mid-stream"],"requires":["API key from Anthropic or OpenRouter","HTTP client library with Server-Sent Events (SSE) support","Streaming parameter enabled in API request (stream=true)","Error handling for connection drops and partial responses"],"input_types":["text (user request)","image (when using multimodal capability)"],"output_types":["text (streamed token-by-token via SSE)","structured events (JSON-wrapped tokens with metadata)"],"categories":["text-generation-language","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-anthropic-claude-3-haiku__cap_6","uri":"capability://automation.workflow.batch.processing.api.for.cost.optimized.high.volume.inference","name":"batch processing api for cost-optimized high-volume inference","description":"Claude 3 Haiku supports batch processing through Anthropic's Batch API, where multiple requests are submitted together and processed asynchronously with a 50% cost discount compared to standard API pricing. Batches are queued and processed during off-peak hours, typically completing within 24 hours. The implementation uses JSONL format for batch submission and provides webhook callbacks or polling for result retrieval.","intents":["I need to process thousands of requests (classification, extraction, summarization) at minimal cost","I want to run overnight batch jobs that don't require real-time responses","I need to analyze large datasets where latency is not a constraint","I want to reduce infrastructure costs for non-interactive workloads"],"best_for":["data teams processing large datasets for analysis or labeling","organizations running nightly ETL pipelines that transform or enrich data","builders creating content processing workflows (summarization, tagging, classification at scale)","teams with cost-sensitive workloads where latency is not critical"],"limitations":["Batch processing introduces 1-24 hour latency; not suitable for real-time or interactive applications","No streaming support in batch mode — responses are returned as complete text","Batch size limits (typically 10,000-100,000 requests per batch) require chunking for very large datasets","Failed requests in a batch must be resubmitted; no built-in retry logic within a batch"],"requires":["API key from Anthropic or OpenRouter","Batch requests formatted as JSONL (JSON Lines) with specific structure","Ability to wait 1-24 hours for results or implement polling/webhook handling","Storage for batch results (typically returned as JSONL file)"],"input_types":["JSONL (batch request format with text and image inputs)"],"output_types":["JSONL (batch results with text responses, structured data, or error messages)"],"categories":["automation-workflow","data-processing-analysis"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-anthropic-claude-3-haiku__cap_7","uri":"capability://image.visual.vision.based.document.and.table.extraction.with.structured.output","name":"vision-based document and table extraction with structured output","description":"Claude 3 Haiku can analyze images of documents, forms, and tables, extracting structured data and converting them to JSON, CSV, or markdown formats. The model uses its vision encoding to understand spatial relationships, text layout, and table structure, then generates structured output that preserves the document's organization. This enables automated document processing without OCR preprocessing or custom layout analysis.","intents":["I need to extract data from scanned documents, invoices, or forms without manual data entry","I want to convert tables in images to structured formats (JSON, CSV) for analysis","I need to process handwritten or poorly scanned documents where traditional OCR fails","I want to understand document structure and relationships between fields without explicit layout parsing"],"best_for":["teams automating document processing workflows (invoice processing, form extraction, data entry)","organizations digitizing paper records or legacy documents","builders creating accessibility tools that convert visual documents to structured formats","data teams extracting information from research papers, reports, or technical documentation"],"limitations":["Accuracy degrades on very low-resolution images (<300 DPI); high-quality scans (300+ DPI) are recommended","Handwriting recognition is less reliable than printed text; cursive or unusual handwriting may be misread","Complex multi-page documents require separate image submissions per page; no built-in document assembly","Structured output format must be specified in the prompt; no automatic schema inference from document type"],"requires":["API key from Anthropic or OpenRouter","Images in JPEG, PNG, GIF, or WebP format","Clear prompting about desired output format (JSON schema, CSV structure, etc.)","Post-processing to validate extracted data against expected schema"],"input_types":["image (document, form, table, or handwritten content)"],"output_types":["JSON (structured extracted data)","CSV (tabular data)","markdown (formatted text with structure)","plain text (with explicit formatting)"],"categories":["image-visual","data-processing-analysis"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-anthropic-claude-3-haiku__cap_8","uri":"capability://code.generation.editing.code.analysis.and.generation.with.multi.language.support","name":"code analysis and generation with multi-language support","description":"Claude 3 Haiku can analyze, generate, and refactor code across 40+ programming languages including Python, JavaScript, Java, C++, Go, Rust, and more. The model understands syntax, semantics, and common patterns for each language, enabling tasks like bug detection, optimization suggestions, and idiomatic code generation. Code understanding is achieved through training on diverse codebases rather than language-specific parsing, enabling the model to handle edge cases and novel patterns.","intents":["I need to generate boilerplate code or implement specific algorithms quickly","I want to review code for bugs, security issues, or performance problems","I need to refactor code to improve readability or performance","I want to translate code between languages or understand unfamiliar code"],"best_for":["developers using Claude as a coding assistant for rapid prototyping","teams automating code review processes or security scanning","builders creating educational tools that explain code or generate examples","organizations migrating codebases between languages or frameworks"],"limitations":["Code generation accuracy varies by language; less common languages (Cobol, Fortran) are less reliable than Python or JavaScript","Large codebases (>50K lines) may exceed context limits; chunking is required for full-codebase analysis","No execution environment; generated code must be tested separately before deployment","Security analysis is heuristic-based; subtle vulnerabilities may be missed, and false positives are common"],"requires":["API key from Anthropic or OpenRouter","Code formatted as text (plain text, markdown code blocks, or syntax-highlighted formats)","Clear prompting about the desired task (generation, review, refactoring, etc.)","Testing infrastructure to validate generated code before deployment"],"input_types":["text (code snippets or full files)","image (screenshots of code, though text is preferred)"],"output_types":["code (generated or refactored code)","text (analysis, suggestions, explanations)","structured data (bug reports, optimization recommendations)"],"categories":["code-generation-editing","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"openrouter-anthropic-claude-3-haiku__cap_9","uri":"capability://text.generation.language.multilingual.text.generation.and.translation.with.cultural.context","name":"multilingual text generation and translation with cultural context","description":"Claude 3 Haiku supports text generation and translation across 50+ languages, maintaining semantic meaning and cultural appropriateness. The model understands language-specific idioms, formality levels, and cultural context, enabling more natural translations than word-for-word approaches. Translation is achieved through the model's general language understanding rather than specialized translation modules, enabling it to handle domain-specific terminology and context-dependent meaning.","intents":["I need to translate content between languages while preserving tone and cultural nuance","I want to generate content in multiple languages from a single prompt","I need to localize applications or content for different markets","I want to understand or summarize content in languages I don't speak"],"best_for":["teams localizing applications or content for international markets","organizations with multilingual customer bases needing content generation","developers building translation features into applications","content creators adapting materials for different linguistic and cultural contexts"],"limitations":["Translation quality varies significantly by language pair; high-resource pairs (English-Spanish) are more accurate than low-resource pairs (English-Icelandic)","Cultural context understanding is probabilistic; some idioms or cultural references may be misinterpreted","No specialized domain knowledge for technical translation; medical, legal, or highly specialized content may require human review","Language detection is automatic but may fail on code-mixed text or very short inputs"],"requires":["API key from Anthropic or OpenRouter","Clear specification of source and target languages in the prompt","UTF-8 encoding support for non-Latin scripts","Human review process for high-stakes translations (legal, medical, marketing)"],"input_types":["text (in any supported language)"],"output_types":["text (translated or generated in target language)","structured data (with language metadata)"],"categories":["text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":26,"verified":false,"data_access_risk":"high","permissions":["API key from Anthropic or OpenRouter","Images in JPEG, PNG, GIF, or WebP format","Base64 encoding for image transmission via API","HTTP/2 capable client for efficient multipart request handling","Network connectivity for API calls (no local inference without custom deployment)","Minimum 1MB/s bandwidth for optimal streaming performance","HTTP/1.1 or HTTP/2 capable client library","Well-chosen examples that clearly demonstrate the desired pattern","Clear task description or system prompt explaining the pattern","Validation data to assess whether few-shot learning is sufficient for the task"],"failure_modes":["Image resolution is limited to ~1568x1568 pixels; larger images are downsampled, potentially losing fine detail","No video frame extraction — must provide individual image frames as separate inputs","Image understanding latency adds ~100-200ms vs text-only inference due to vision encoding overhead","Cannot generate, edit, or manipulate images — vision is read-only","Smaller effective context window (200K tokens) compared to Claude 3 Sonnet (200K) — no advantage here, but reasoning depth is shallower","Lower accuracy on complex multi-step reasoning tasks; performance degrades on problems requiring >5 reasoning steps","Cannot reliably handle very long documents (>50K tokens) with high accuracy due to attention pattern limitations","Knowledge cutoff date limits real-time information access; requires external retrieval for current events","Few-shot learning performance plateaus around 5-10 examples; adding more examples doesn't significantly improve accuracy and wastes context tokens","Complex tasks requiring deep domain knowledge may not be learnable from examples alone; fine-tuning is more effective for specialized domains","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.05,"quality":0.47,"ecosystem":0.37,"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":"2026-05-24T12:16:24.483Z","last_scraped_at":"2026-05-03T15:20:45.777Z","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=anthropic-claude-3-haiku","compare_url":"https://unfragile.ai/compare?artifact=anthropic-claude-3-haiku"}},"signature":"bKHIH7jjzIAQfDfKbiipsO6V83Pxsy5QhyA0ztJ+znw4Qbx4ClO0TbZPXYVbwmmER5qsoNF/9UjGsc5WL1gqDQ==","signedAt":"2026-06-20T21:38:42.964Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/anthropic-claude-3-haiku","artifact":"https://unfragile.ai/anthropic-claude-3-haiku","verify":"https://unfragile.ai/api/v1/verify?slug=anthropic-claude-3-haiku","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"}}