{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"mixtral-8x22b","slug":"mixtral-8x22b","name":"Mixtral 8x22B","type":"model","url":"https://mistral.ai/news/mixtral-8x22b/","page_url":"https://unfragile.ai/mixtral-8x22b","categories":["model-training"],"tags":[],"pricing":{"model":"free","free":true,"starting_price":null},"status":"active","verified":false},"capabilities":[{"id":"mixtral-8x22b__cap_0","uri":"capability://text.generation.language.sparse.mixture.of.experts.text.generation","name":"sparse-mixture-of-experts-text-generation","description":"Generates text using a sparse mixture-of-experts architecture with 8 experts of 22B parameters each, activating only 2 experts per token for 44B active parameters. This sparse activation pattern reduces computational cost compared to dense models while maintaining 176B total parameter capacity. The routing mechanism dynamically selects which 2 experts process each token based on learned gating functions, enabling efficient inference on consumer hardware.","intents":["Generate high-quality text completions with lower inference latency than dense 70B models","Build applications requiring fast token generation without sacrificing model quality","Deploy language models on resource-constrained infrastructure while maintaining reasoning capability","Process long documents up to 64K tokens without context truncation"],"best_for":["Teams building production LLM applications prioritizing inference speed and cost efficiency","Developers deploying open-source models on limited GPU VRAM (sparse activation reduces memory footprint vs dense equivalents)","Organizations requiring Apache 2.0 licensed models for commercial applications without licensing restrictions"],"limitations":["Sparse activation requires inference frameworks optimized for mixture-of-experts (vLLM, TensorRT-LLM support confirmed; broader framework compatibility unknown)","No quantization format availability documented (GGUF, int8, int4 support status unknown)","Specific throughput metrics (tokens/second) not published; claimed faster than dense 70B but exact speedup undefined","Expert load balancing may cause uneven GPU utilization on multi-GPU setups without specialized scheduling"],"requires":["GPU with sufficient VRAM for 44B active parameters (exact requirement not documented; estimate 80-100GB for full precision)","Inference framework supporting sparse mixture-of-experts (vLLM recommended; TensorRT-LLM compatibility unknown)","API key for Mistral la Plateforme OR self-hosted deployment infrastructure","64K token context window support in downstream application"],"input_types":["text prompts","multi-turn conversation history","code snippets for in-context learning"],"output_types":["text generation","structured text with function calling constraints"],"categories":["text-generation-language","sparse-inference"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__cap_1","uri":"capability://tool.use.integration.native.function.calling.with.constrained.output","name":"native-function-calling-with-constrained-output","description":"Supports structured function calling through native integration with Mistral's constrained output mode on la Plateforme, enabling the model to generate function calls in a schema-compliant format without hallucinating invalid function names or parameters. The model learns during training to recognize function schemas and produce valid JSON-formatted function calls that downstream systems can parse and execute deterministically.","intents":["Build agentic systems that reliably call external tools without parsing errors or invalid function invocations","Create API wrappers where the model must generate valid function calls matching predefined schemas","Implement tool-use workflows requiring guaranteed schema compliance without post-processing validation"],"best_for":["Developers building AI agents on Mistral la Plateforme requiring deterministic tool calling","Teams implementing function-calling workflows where schema validation is critical","Applications where invalid function calls would cause downstream system failures"],"limitations":["Constrained output mode only available on la Plateforme; self-hosted deployments may not support this feature","Function schema complexity limits unknown — no documentation on maximum schema size or nesting depth","Requires explicit schema definition per function; no automatic schema inference from Python type hints or OpenAPI specs documented","Performance overhead of constrained decoding not quantified; may add latency vs unconstrained generation"],"requires":["Mistral API account with la Plateforme access","Function schemas defined in JSON Schema format (exact format specification not documented)","Integration with Mistral's constrained output API endpoint"],"input_types":["text prompts with function schema definitions","multi-turn conversations with tool descriptions"],"output_types":["JSON-formatted function calls matching provided schema","structured function name + parameters pairs"],"categories":["tool-use-integration","structured-output"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__cap_10","uri":"capability://text.generation.language.instruction.tuned.variant.for.chat.and.tasks","name":"instruction-tuned-variant-for-chat-and-tasks","description":"An instruction-tuned variant of Mixtral 8x22B is available, optimized for following user instructions, chat interactions, and task-specific prompts. This variant shows improved performance on mathematical reasoning (90.8% GSM8K, 44.6% MATH) and likely better instruction-following compared to the base model. The instruction-tuning process teaches the model to recognize task descriptions and generate appropriate responses aligned with user intent.","intents":["Build chatbots and conversational AI systems that follow user instructions accurately","Create task-specific applications (summarization, translation, Q&A) with better instruction adherence","Implement assistant-like systems that understand and execute complex multi-step instructions"],"best_for":["Conversational AI and chatbot applications requiring instruction-following capability","Task-specific systems (summarization, translation, code generation) where instruction clarity is important","Applications where user intent alignment is critical for user satisfaction"],"limitations":["Instruction-tuning approach and dataset not documented — unclear what instructions the model was trained on","No benchmark comparison between base and instruction-tuned variants (except math tasks) — improvement on general tasks unknown","Instruction-tuning may reduce diversity or creativity in outputs — no documentation of this trade-off","Fine-tuning guidance for custom instructions not provided — unclear how to adapt the model for domain-specific tasks"],"requires":["Use the instruction-tuned variant (not base model) for best results","Clear, well-formatted instructions in prompts","Task descriptions that align with instruction-tuning distribution"],"input_types":["natural language instructions","task descriptions","multi-turn conversation history","structured prompts with task context"],"output_types":["task-specific responses","chat completions","structured outputs aligned with instructions"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__cap_11","uri":"capability://text.generation.language.mmlu.benchmark.performance.at.77.8.accuracy","name":"mmlu benchmark performance at 77.8% accuracy","description":"Achieves 77.8% accuracy on the Massive Multitask Language Understanding (MMLU) benchmark, a comprehensive evaluation of knowledge across 57 diverse subjects including STEM, humanities, and social sciences. This benchmark score indicates broad knowledge coverage and reasoning capability across multiple domains. The score positions Mixtral 8x22B as a capable general-purpose model suitable for knowledge-intensive tasks, though specific subject-level performance breakdown is not provided.","intents":["I need to evaluate Mixtral 8x22B's general knowledge and reasoning capabilities","I want to understand how this model compares to other open-weight models on standardized benchmarks","I need to assess whether the model is suitable for knowledge-intensive applications"],"best_for":["teams evaluating open-weight models for knowledge-intensive applications","researchers comparing model capabilities across benchmarks","organizations assessing model suitability for question-answering or knowledge retrieval tasks"],"limitations":["MMLU score of 77.8% is lower than proprietary models (GPT-4: 86.4%, Claude 3 Opus: 88.7%), indicating capability gaps","Subject-level performance breakdown not provided; unclear which domains the model excels in or struggles with","MMLU is a multiple-choice benchmark; performance on open-ended knowledge tasks may differ","No comparison with other open-weight models' MMLU scores provided in documentation"],"requires":["Understanding of MMLU benchmark structure and evaluation methodology","Awareness that benchmark performance may not translate to real-world task performance"],"input_types":["MMLU multiple-choice questions"],"output_types":["multiple-choice answers","knowledge-based responses"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__cap_2","uri":"capability://text.generation.language.multilingual.text.generation.across.five.languages","name":"multilingual-text-generation-across-five-languages","description":"Generates fluent text in English, French, Italian, German, and Spanish with native language understanding trained into the model weights. The model demonstrates strong cross-lingual performance on benchmarks like MMLU and HellaSwag, outperforming Llama 2 70B on multilingual variants. Language selection is implicit in the input prompt; no explicit language-switching mechanism is required.","intents":["Build multilingual applications serving European markets without separate language-specific models","Generate content in multiple languages from a single model, reducing deployment complexity","Provide customer support chatbots with native fluency across English, French, German, Italian, and Spanish"],"best_for":["European SaaS companies requiring multilingual support without model multiplication","Content generation platforms serving multiple language markets","Teams building chatbots or customer service systems for French, German, Italian, or Spanish-speaking regions"],"limitations":["Only 5 languages officially supported (English, French, Italian, German, Spanish); other languages not documented as supported","Multilingual performance scores not provided in detail — only stated as 'outperforms Llama 2 70B' without specific benchmark numbers","Code-switching (mixing languages in single prompt) behavior not documented","No language-specific fine-tuning guidance provided; unclear if model maintains language separation or exhibits cross-lingual interference"],"requires":["Input text in one of the five supported languages","No explicit language specification required; model infers from prompt context"],"input_types":["text prompts in English, French, Italian, German, or Spanish","multilingual conversation histories"],"output_types":["text generation in the same language as input","code-switched output if input mixes languages (behavior undefined)"],"categories":["text-generation-language","multilingual"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__cap_3","uri":"capability://text.generation.language.mathematical.reasoning.with.instruction.tuning","name":"mathematical-reasoning-with-instruction-tuning","description":"The instructed version of Mixtral 8x22B achieves 90.8% on GSM8K (grade-school math with majority voting over 8 samples) and 44.6% on MATH (competition-level mathematics with majority voting over 4 samples) through instruction-tuning that teaches the model to decompose mathematical problems into step-by-step reasoning chains. The model learns to recognize mathematical operators, maintain numerical precision, and apply algebraic transformations correctly.","intents":["Build educational tutoring systems that solve math problems with step-by-step explanations","Create applications requiring reliable arithmetic and algebraic reasoning","Implement homework assistance tools with strong performance on grade-school and intermediate math"],"best_for":["EdTech platforms requiring strong grade-school math reasoning (GSM8K performance: 90.8%)","Applications where intermediate math accuracy is acceptable but competition-level math is not required","Teams building tutoring systems where step-by-step reasoning is more important than final answer correctness"],"limitations":["Competition-level mathematics performance is limited (44.6% on MATH benchmark) — not suitable for advanced STEM applications","Majority voting results require generating multiple samples (8 for GSM8K, 4 for MATH), increasing inference cost and latency","Numerical precision limits unknown — no documentation on handling very large numbers, floating-point edge cases, or symbolic math","Only instructed version achieves these scores; base model performance on math tasks not documented"],"requires":["Instructed version of Mixtral 8x22B (not base model)","For best results: majority voting over 4-8 samples (increases latency and cost proportionally)","Math problems formatted as natural language text (no LaTeX or symbolic math format documented)"],"input_types":["natural language math problems","word problems","arithmetic and algebraic expressions in text form"],"output_types":["step-by-step reasoning chains","numerical answers","algebraic solutions"],"categories":["text-generation-language","reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__cap_4","uri":"capability://text.generation.language.64k.token.context.window.for.long.document.processing","name":"64k-token-context-window-for-long-document-processing","description":"Supports a native 64K token context window, enabling the model to process documents, conversations, and code repositories up to approximately 48,000 words without truncation or sliding-window approximations. The context window is implemented as a standard transformer attention mechanism scaled to 64K positions, allowing the model to maintain coherence across long-range dependencies and reference information from document beginnings in later generations.","intents":["Process entire documents, research papers, or code repositories in a single context without chunking","Build long-form content generation systems that maintain consistency across thousands of tokens","Implement document analysis and summarization for lengthy texts without information loss from truncation"],"best_for":["Document analysis platforms processing research papers, legal contracts, or technical specifications","Code analysis tools requiring full repository context for refactoring or bug detection","Long-form content generation systems (e.g., book chapters, technical documentation)"],"limitations":["64K token hard limit — documents exceeding ~48,000 words must be chunked or summarized before input","No documented performance degradation at context boundaries, but transformer attention complexity is O(n²) — latency increases quadratically with context length","Inference cost scales linearly with context length; processing 64K tokens costs ~1.5x more than processing 32K tokens","Long-context performance on retrieval and reasoning tasks not benchmarked — no documentation of accuracy degradation at 50K+ tokens"],"requires":["Input text or code up to 64K tokens (approximately 48,000 words)","Inference infrastructure supporting 64K sequence length (may require specific vLLM or TensorRT-LLM configurations)","Sufficient GPU VRAM for KV cache storage (64K × 2 × hidden_dim × batch_size; exact requirement not documented)"],"input_types":["long-form text documents","code repositories or files","multi-turn conversation histories","concatenated documents with separator tokens"],"output_types":["text generation conditioned on full context","summaries or analyses of long documents","code generation with full repository awareness"],"categories":["text-generation-language","memory-knowledge"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__cap_5","uri":"capability://code.generation.editing.code.generation.with.sparse.activation","name":"code-generation-with-sparse-activation","description":"Generates code across multiple programming languages using the sparse mixture-of-experts architecture, where expert routing dynamically selects relevant experts for code-specific patterns. The model learns to recognize syntax, semantics, and common code patterns during training, enabling it to complete functions, refactor code, and generate bug fixes. Specific code language support and performance metrics (HumanEval, MBPP) are not detailed in available documentation.","intents":["Generate code completions and function implementations from natural language descriptions","Refactor or optimize existing code snippets","Implement code review and bug detection by generating corrected versions"],"best_for":["Developers using IDE plugins or API-based code completion tools","Teams building code generation features into development tools","Organizations requiring open-source code models without licensing restrictions"],"limitations":["Specific code language support not documented — unclear which languages are optimized vs supported","Code generation benchmarks (HumanEval, MBPP) shown in documentation but specific scores not provided","No documentation on code-specific fine-tuning or instruction-tuning approach","Sparse activation may cause inconsistent performance on code patterns that don't align with expert specialization"],"requires":["Code snippets or natural language descriptions of desired code","Target programming language specified in prompt (implicit or explicit)"],"input_types":["natural language function descriptions","code snippets for completion or refactoring","bug descriptions for fix generation"],"output_types":["code completions","full function implementations","refactored code","bug fixes"],"categories":["code-generation-editing","sparse-inference"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__cap_6","uri":"capability://automation.workflow.apache.2.0.licensed.open.source.deployment","name":"apache-2-0-licensed-open-source-deployment","description":"Released under Apache 2.0 license, enabling unrestricted commercial use, modification, and redistribution of model weights and code. The model is available for download and self-hosting without licensing fees or usage restrictions, making it suitable for proprietary applications and commercial products. License compliance requires only attribution and license inclusion in derivative works.","intents":["Deploy models in proprietary commercial applications without licensing fees or usage restrictions","Fine-tune and redistribute modified versions of the model for specific use cases","Build closed-source products using open-source model weights without legal constraints"],"best_for":["Commercial companies requiring open-source models for proprietary products","Startups avoiding licensing costs and vendor lock-in of closed-source models","Organizations with strict IP requirements or need for model customization"],"limitations":["Apache 2.0 requires attribution and license inclusion in derivative works — not true 'no strings attached' licensing","No warranty or liability protection — model is provided 'as-is' without guarantees","No commercial support or SLA from Mistral AI included with open-source release (support available via paid API)","Self-hosting requires infrastructure management and optimization (no managed service included)"],"requires":["Acknowledgment of Apache 2.0 license in product documentation or code","Inclusion of license text in distributed software","No restrictions on commercial use, but attribution required"],"input_types":["model weights (available for download)","source code for fine-tuning and deployment"],"output_types":["modified model weights","fine-tuned versions","proprietary applications using the model"],"categories":["automation-workflow","licensing"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__cap_7","uri":"capability://automation.workflow.mistral.la.plateforme.api.deployment","name":"mistral-la-plateforme-api-deployment","description":"Available for deployment on Mistral's managed API platform (la Plateforme), providing hosted inference without self-hosting infrastructure. The platform handles model serving, scaling, and optimization, exposing the model through REST API endpoints. Pricing is consumption-based (per-token), and the platform includes features like constrained output mode for function calling and automatic batching for throughput optimization.","intents":["Deploy Mixtral 8x22B without managing inference infrastructure or GPU resources","Build applications with automatic scaling and high availability without DevOps overhead","Access the model through standard REST APIs without custom deployment code"],"best_for":["Startups and small teams without DevOps infrastructure for model deployment","Applications requiring automatic scaling and high availability","Developers prioritizing time-to-market over infrastructure control"],"limitations":["Vendor lock-in to Mistral's infrastructure — switching to self-hosted or competitor APIs requires code changes","API pricing not documented in artifact — exact cost per token unknown, preventing cost comparison vs self-hosting","Latency and throughput depend on Mistral's infrastructure and network conditions — no SLA or performance guarantees documented","Data privacy concerns — prompts and outputs are sent to Mistral's servers (may violate compliance requirements for sensitive data)","Rate limiting and quota restrictions not documented — unclear if there are usage caps or burst limits"],"requires":["Mistral API account and authentication credentials","API key for request authentication","Network connectivity to Mistral's API endpoints","Pricing agreement and payment method for consumption-based billing"],"input_types":["text prompts via REST API","JSON-formatted requests with model parameters"],"output_types":["text completions via REST API responses","streaming responses (if supported)","structured function calls with constrained output mode"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__cap_8","uri":"capability://text.generation.language.general.knowledge.reasoning.on.mmlu.benchmark","name":"general-knowledge-reasoning-on-mmlu-benchmark","description":"Achieves 77.8% accuracy on the MMLU (Massive Multitask Language Understanding) benchmark, which tests knowledge across 57 diverse subjects including STEM, humanities, and professional domains. This benchmark measures the model's ability to reason about factual knowledge, apply domain-specific concepts, and select correct answers from multiple choices. The score positions Mixtral 8x22B as a capable general-knowledge model suitable for knowledge-intensive applications.","intents":["Build question-answering systems that test knowledge across diverse domains","Create educational assessment tools that evaluate understanding of multiple subjects","Implement knowledge-based chatbots that can reason about facts and concepts"],"best_for":["Educational platforms requiring general knowledge reasoning across multiple subjects","Trivia and quiz applications requiring broad knowledge coverage","Knowledge-intensive applications where factual accuracy is important"],"limitations":["77.8% MMLU accuracy means ~22% error rate — not suitable for applications requiring near-perfect accuracy","MMLU tests multiple-choice reasoning, not open-ended knowledge generation — performance on free-form knowledge questions unknown","No breakdown by subject domain provided — unclear which subjects the model excels at vs struggles with","Hallucination rate and factual accuracy on knowledge outside MMLU scope not documented"],"requires":["Knowledge questions formatted as multiple-choice (MMLU format) or open-ended text","Domain knowledge within the 57 subjects covered by MMLU"],"input_types":["multiple-choice questions","knowledge-based prompts","factual queries"],"output_types":["answer selections from multiple choices","factual explanations","reasoning chains supporting answers"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__cap_9","uri":"capability://automation.workflow.self.hosted.deployment.with.apache.2.0.weights","name":"self-hosted-deployment-with-apache-2-0-weights","description":"Model weights are available for download and self-hosting on custom infrastructure, enabling organizations to run Mixtral 8x22B on their own hardware without relying on Mistral's managed API. Self-hosting requires compatible inference frameworks (vLLM, TensorRT-LLM, or similar) and sufficient GPU resources to load and run the sparse mixture-of-experts model. This approach provides full control over data privacy, latency, and cost structure.","intents":["Deploy the model on private infrastructure for data privacy and compliance requirements","Optimize inference latency and throughput for specific hardware configurations","Avoid vendor lock-in and API costs by running the model locally"],"best_for":["Organizations with strict data privacy or compliance requirements (HIPAA, GDPR, etc.)","Teams with existing GPU infrastructure and DevOps expertise","Applications requiring sub-100ms latency where API round-trip time is unacceptable"],"limitations":["GPU VRAM requirements not documented — exact memory footprint for 44B active parameters unknown (estimate: 80-100GB for full precision)","Inference framework compatibility not fully documented — vLLM support likely but TensorRT-LLM, ONNX, and other frameworks unknown","No quantization formats documented (GGUF, int8, int4) — unclear if lower-precision versions are available","Sparse activation optimization requires specialized inference code — standard dense model frameworks may not efficiently utilize expert routing","Operational overhead: monitoring, scaling, security patching, and model updates are customer responsibility"],"requires":["GPU with sufficient VRAM (estimated 80-100GB for full precision; exact requirement unknown)","Inference framework supporting sparse mixture-of-experts (vLLM recommended; others unknown)","Model weights downloaded from Mistral's repository (format: likely safetensors or PyTorch)","Infrastructure for serving (Kubernetes, Docker, or bare metal)","DevOps expertise for deployment, monitoring, and scaling"],"input_types":["model weights in safetensors or PyTorch format","text prompts via local API or library calls"],"output_types":["text completions","structured outputs with constrained decoding (if framework supports it)"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x22b__headline","uri":"capability://text.generation.language.mixture.of.experts.model.for.multilingual.text.generation.and.coding.tasks","name":"mixture-of-experts model for multilingual text generation and coding tasks","description":"Mixtral 8x22B is a state-of-the-art Mixture-of-Experts model designed for efficient multilingual text generation and coding tasks, leveraging 176 billion parameters with 44 billion active parameters for optimal performance.","intents":["best mixture-of-experts model","multilingual model for coding tasks","efficient text generation model","AI model for function calling","top-performing model on MMLU benchmark"],"best_for":["multilingual applications","coding and mathematics tasks"],"limitations":[],"requires":[],"input_types":[],"output_types":[],"categories":["text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":57,"verified":false,"data_access_risk":"low","permissions":["GPU with sufficient VRAM for 44B active parameters (exact requirement not documented; estimate 80-100GB for full precision)","Inference framework supporting sparse mixture-of-experts (vLLM recommended; TensorRT-LLM compatibility unknown)","API key for Mistral la Plateforme OR self-hosted deployment infrastructure","64K token context window support in downstream application","Mistral API account with la Plateforme access","Function schemas defined in JSON Schema format (exact format specification not documented)","Integration with Mistral's constrained output API endpoint","Use the instruction-tuned variant (not base model) for best results","Clear, well-formatted instructions in prompts","Task descriptions that align with instruction-tuning distribution"],"failure_modes":["Sparse activation requires inference frameworks optimized for mixture-of-experts (vLLM, TensorRT-LLM support confirmed; broader framework compatibility unknown)","No quantization format availability documented (GGUF, int8, int4 support status unknown)","Specific throughput metrics (tokens/second) not published; claimed faster than dense 70B but exact speedup undefined","Expert load balancing may cause uneven GPU utilization on multi-GPU setups without specialized scheduling","Constrained output mode only available on la Plateforme; self-hosted deployments may not support this feature","Function schema complexity limits unknown — no documentation on maximum schema size or nesting depth","Requires explicit schema definition per function; no automatic schema inference from Python type hints or OpenAPI specs documented","Performance overhead of constrained decoding not quantified; may add latency vs unconstrained generation","Instruction-tuning approach and dataset not documented — unclear what instructions the model was trained on","No benchmark comparison between base and instruction-tuned variants (except math tasks) — improvement on general tasks unknown","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.7,"quality":0.9,"ecosystem":0.3,"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":"2026-05-24T12:16:23.328Z","last_scraped_at":null,"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=mixtral-8x22b","compare_url":"https://unfragile.ai/compare?artifact=mixtral-8x22b"}},"signature":"mGd0zlKJu5mhrPs0qmBSUmKQxGkO2zEq4Y+llJl9kpip18n56MGP0ywBIIi3shi3JDbP8vr3/z8A3DIySOhuBw==","signedAt":"2026-06-21T07:52:52.713Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/mixtral-8x22b","artifact":"https://unfragile.ai/mixtral-8x22b","verify":"https://unfragile.ai/api/v1/verify?slug=mixtral-8x22b","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"}}