{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"mixtral-8x7b","slug":"mixtral-8x7b","name":"Mixtral 8x7B","type":"model","url":"https://mistral.ai/news/mixtral-of-experts","page_url":"https://unfragile.ai/mixtral-8x7b","categories":["model-training"],"tags":[],"pricing":{"model":"free","free":true,"starting_price":null},"status":"active","verified":false},"capabilities":[{"id":"mixtral-8x7b__cap_0","uri":"capability://text.generation.language.sparse.mixture.of.experts.token.routing","name":"sparse-mixture-of-experts-token-routing","description":"Routes each input token through exactly 2 of 8 expert networks per transformer layer using a learned router network, activating only 12.9B of 46.7B total parameters per forward pass. The router makes independent routing decisions per token per layer, with expert outputs combined additively. This sparse activation pattern enables inference throughput equivalent to a 12.9B dense model while maintaining GPT-3.5-level performance across benchmarks.","intents":["Deploy a model with GPT-3.5-level capabilities but 6x faster inference than Llama 2 70B","Run inference on consumer-grade hardware with the performance of a 70B-parameter model","Understand how sparse routing mechanisms reduce computational cost without proportional performance loss","Optimize inference cost and latency for production language model deployments"],"best_for":["ML engineers optimizing inference cost and latency for production systems","Teams deploying open-source models where inference speed directly impacts operational costs","Researchers studying sparse mixture-of-experts architectures and token-level routing"],"limitations":["Router network design and load-balancing strategy not publicly documented; potential for uneven expert utilization or routing instability under adversarial inputs","Sparse activation introduces routing overhead (~5-10% estimated) that is not quantified in official benchmarks","No documented failure modes for routing decisions or expert saturation scenarios","Routing decisions are per-token per-layer, creating potential for token-specific bottlenecks if certain experts become overloaded"],"requires":["CUDA-capable GPU with sufficient VRAM (exact requirements not specified; estimated 24GB+ for full precision inference)","vLLM with Megablocks CUDA kernels for optimized sparse inference","Python 3.8+ for inference frameworks"],"input_types":["text tokens (up to 32,768 context window)"],"output_types":["text tokens (generated via sparse expert routing)"],"categories":["text-generation-language","model-architecture"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__cap_1","uri":"capability://text.generation.language.gpt.35.level.general.language.generation","name":"gpt-35-level-general-language-generation","description":"Generates coherent, contextually-aware text across general-purpose language tasks by applying transformer decoder architecture with 32K token context window. The model was trained on open web data and achieves performance parity with GPT-3.5 on standard benchmarks (MMLU, HellaSwag, TruthfulQA, Winogrande, GSM8K, MATH, HumanEval) while maintaining lower computational cost through sparse routing. Supports both base and instruction-tuned variants, with the Instruct variant fine-tuned via supervised fine-tuning (SFT) and Direct Preference Optimization (DPO).","intents":["Generate high-quality text for content creation, summarization, and general Q&A without relying on proprietary APIs","Self-host a model with GPT-3.5-equivalent capabilities under Apache 2.0 license for full control and privacy","Benchmark open-source model performance against commercial alternatives using standard evaluation metrics","Use instruction-following capabilities for chatbot and assistant applications via the Instruct variant"],"best_for":["Organizations requiring GPT-3.5-level performance with open-source licensing and self-hosting capability","Developers building privacy-critical applications where model weights and inference must remain on-premises","Teams evaluating open-source alternatives to commercial LLMs for cost optimization"],"limitations":["Base model has no built-in safety guardrails; requires explicit prompting or preference tuning to refuse harmful requests","Context window hard-limited to 32,768 tokens; cannot process documents longer than this without chunking","Performance on specialized domains (medical, legal, scientific) not explicitly documented; general benchmarks may not reflect domain-specific capability gaps","Hallucination and factual accuracy not quantified; relies on same training data quality as other open-web-trained models"],"requires":["GPU with sufficient VRAM for inference (estimated 24GB+ for full precision, 12GB+ for quantized variants)","vLLM or similar inference framework for efficient serving","Python 3.8+ for integration"],"input_types":["text prompts (up to 32,768 tokens)"],"output_types":["generated text (variable length, typically 1-2K tokens per generation)"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__cap_10","uri":"capability://data.processing.analysis.benchmark.evaluation.across.standard.metrics","name":"benchmark-evaluation-across-standard-metrics","description":"Evaluated across standard language model benchmarks including MMLU (knowledge), HellaSwag (common sense reasoning), TruthfulQA (factuality), Winogrande (coreference resolution), GSM8K (math), MATH (advanced math), and HumanEval (code generation). Results demonstrate performance parity with GPT-3.5 on most benchmarks, with specific scores provided for MT-Bench (8.30 for Instruct variant). Benchmark evaluation enables quantitative comparison with other models and verification of capability claims.","intents":["Evaluate Mixtral performance against GPT-3.5 and other open-source models using standard benchmarks","Verify capability claims through independent benchmark evaluation","Compare performance across different task categories (knowledge, reasoning, math, code)","Make informed model selection decisions based on quantitative benchmark results"],"best_for":["ML engineers and researchers evaluating model capabilities through standard benchmarks","Teams making model selection decisions based on quantitative performance metrics","Organizations benchmarking open-source models against commercial alternatives"],"limitations":["Benchmark scores reflect performance on specific evaluation datasets; real-world performance may differ","Benchmarks measure narrow capabilities (MMLU knowledge, HumanEval code); broader capability assessment requires additional evaluation","Specific benchmark scores not provided in available documentation for most benchmarks; only comparative claims ('matches or outperforms GPT-3.5') documented","MT-Bench score of 8.30 is single metric; performance on specialized instruction-following tasks not separately quantified"],"requires":["Benchmark datasets (MMLU, HellaSwag, TruthfulQA, Winogrande, GSM8K, MATH, HumanEval, MT-Bench)","Evaluation framework (e.g., lm-evaluation-harness)","GPU with sufficient VRAM for inference","Python 3.8+"],"input_types":["benchmark evaluation prompts and test cases"],"output_types":["benchmark scores and performance metrics"],"categories":["data-processing-analysis","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__cap_11","uri":"capability://safety.moderation.no.built.in.safety.guardrails.base.model","name":"no-built-in-safety-guardrails-base-model","description":"Base model (non-Instruct variant) has no built-in safety guardrails and will follow any instruction without refusal or content filtering. Safety behavior is not enforced through training or architecture; instead, the model relies on explicit prompting or preference optimization (as in the Instruct variant) to learn refusal behavior. This design choice prioritizes capability and flexibility over safety by default, requiring users to implement safety measures explicitly.","intents":["Use base model for research and evaluation without safety constraints that might interfere with capability assessment","Fine-tune base model for specific domains where default safety guardrails might be inappropriate","Understand model behavior without safety-induced refusals or content filtering","Implement custom safety measures tailored to specific application requirements"],"best_for":["Researchers studying model behavior and safety without built-in constraints","Teams fine-tuning models for domains where default safety guardrails are inappropriate","Organizations implementing custom safety measures aligned with specific requirements"],"limitations":["Base model can be prompted to generate harmful, illegal, or unethical content without refusal","No content filtering or safety checks; requires external moderation or safety layers for production use","Safety behavior must be explicitly implemented through prompting, fine-tuning, or external systems","Instruction-tuned variant (Instruct) has learned safety behavior through DPO, but base model does not"],"requires":["Explicit safety implementation (prompting, fine-tuning, or external moderation)","Understanding of model behavior and potential failure modes","Responsible deployment practices and monitoring"],"input_types":["any text prompt (including harmful or unsafe requests)"],"output_types":["generated text without safety filtering or refusal"],"categories":["safety-moderation","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__cap_2","uri":"capability://code.generation.editing.code.generation.and.completion","name":"code-generation-and-completion","description":"Generates and completes code across multiple programming languages by applying transformer decoder architecture trained on code-inclusive datasets. The model demonstrates strong performance on HumanEval benchmark and supports code generation for tasks ranging from single-function completion to multi-file refactoring. Instruction-tuned variant (Mixtral 8x7B Instruct) provides improved code understanding and explanation capabilities through supervised fine-tuning and preference optimization.","intents":["Generate code snippets and complete partial code implementations across multiple languages","Use as a code assistant for rapid prototyping and boilerplate generation","Integrate code generation into development workflows via self-hosted inference","Evaluate open-source code generation performance against commercial alternatives like GitHub Copilot"],"best_for":["Solo developers and small teams building tools that require code generation without external API dependencies","Organizations with strict data privacy requirements where code cannot be sent to third-party APIs","Researchers benchmarking code generation capabilities of open-source models"],"limitations":["Code generation quality not quantified beyond HumanEval score; performance on domain-specific languages or legacy code patterns unknown","No built-in code execution or validation; generated code must be tested separately","Context window of 32K tokens limits ability to generate code for very large files or multi-file refactoring tasks","No explicit support for IDE integration or real-time code completion like commercial tools; requires custom integration"],"requires":["GPU with 24GB+ VRAM for full-precision inference","vLLM or similar inference framework","Python 3.8+ for integration"],"input_types":["code snippets and natural language descriptions (up to 32,768 tokens)"],"output_types":["generated code in multiple programming languages"],"categories":["code-generation-editing","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__cap_3","uri":"capability://text.generation.language.multilingual.text.generation","name":"multilingual-text-generation","description":"Generates coherent text in English, French, German, Spanish, and Italian through transformer decoder architecture trained on multilingual open web data. The model maintains language-specific performance across supported languages while using the same sparse routing mechanism as English generation. Multilingual performance is documented with benchmark results for each language, though specific scores are not detailed in available documentation.","intents":["Generate content in multiple European languages without requiring separate language-specific models","Build multilingual chatbots and content generation systems with a single model","Evaluate multilingual performance of open-source models against commercial alternatives","Support international applications with consistent model behavior across languages"],"best_for":["Teams building multilingual applications across European markets","Organizations requiring language support without maintaining separate model instances","Developers evaluating multilingual capabilities of open-source models"],"limitations":["Multilingual support limited to 5 languages (English, French, German, Spanish, Italian); no support for non-European languages or non-Latin scripts","Specific multilingual benchmark scores not provided in documentation; relative performance across languages unknown","Code-switching and mixed-language generation behavior not documented","Language-specific fine-tuning or preference optimization not mentioned; multilingual performance relies on base training only"],"requires":["GPU with 24GB+ VRAM","vLLM or similar inference framework","Python 3.8+"],"input_types":["text in English, French, German, Spanish, or Italian (up to 32,768 tokens)"],"output_types":["generated text in the input language"],"categories":["text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__cap_4","uri":"capability://text.generation.language.instruction.following.and.chat","name":"instruction-following-and-chat","description":"Follows natural language instructions and engages in multi-turn conversation through the Mixtral 8x7B Instruct variant, which is fine-tuned via supervised fine-tuning (SFT) and Direct Preference Optimization (DPO). The instruction-tuned variant achieves MT-Bench score of 8.30, positioning it as the best open-source model on this benchmark at release. The model learns to refuse harmful requests and provide helpful, harmless, and honest responses through preference optimization, though safety guardrails are not guaranteed without explicit prompting.","intents":["Build chatbots and conversational AI systems that follow user instructions reliably","Deploy instruction-following models for customer support and Q&A applications","Evaluate instruction-following quality of open-source models using MT-Bench and similar benchmarks","Create assistants that can refuse harmful requests and provide nuanced responses"],"best_for":["Teams building chatbot and conversational AI applications with open-source models","Organizations requiring instruction-following capabilities without proprietary API dependencies","Developers evaluating instruction-tuning effectiveness via MT-Bench and similar benchmarks"],"limitations":["Safety guardrails are learned through preference optimization but not guaranteed; model can still be jailbroken or prompted to produce harmful content","MT-Bench score of 8.30 is a single-metric evaluation; performance on specialized instruction-following tasks (code, math, reasoning) not separately quantified","Instruction-following quality depends on prompt engineering; unclear how robust the model is to adversarial or out-of-distribution instructions","No explicit documentation of refusal behavior or safety boundaries; requires empirical testing for specific use cases"],"requires":["GPU with 24GB+ VRAM","vLLM or similar inference framework","Python 3.8+","Mixtral 8x7B Instruct variant (not base model)"],"input_types":["natural language instructions and multi-turn conversation (up to 32,768 tokens)"],"output_types":["instruction-following responses and conversational text"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__cap_5","uri":"capability://automation.workflow.efficient.inference.via.vllm.megablocks","name":"efficient-inference-via-vllm-megablocks","description":"Enables efficient inference through integration with vLLM framework and Megablocks CUDA kernels, which are specifically optimized for sparse mixture-of-experts computation. The sparse activation pattern (2 of 8 experts per token) is implemented via custom CUDA kernels that avoid computing inactive expert parameters, reducing memory bandwidth and compute requirements. Inference throughput is equivalent to a 12.9B dense model despite 46.7B total parameters, achieving 6x speedup over Llama 2 70B while maintaining equivalent performance.","intents":["Deploy Mixtral with optimized inference performance using vLLM and Megablocks kernels","Reduce inference latency and cost by leveraging sparse activation without performance loss","Understand how custom CUDA kernels enable efficient sparse model inference","Benchmark inference speed improvements from sparse routing vs dense models"],"best_for":["ML engineers optimizing inference performance for production deployments","Teams deploying language models where inference cost directly impacts operational margins","Researchers studying efficient inference techniques for sparse mixture-of-experts models"],"limitations":["Megablocks CUDA kernels require NVIDIA GPU; no support for AMD, Intel, or CPU-only inference","vLLM integration requires specific version compatibility; kernel updates may break inference pipelines","Sparse activation overhead (routing decisions, expert selection) not quantified; actual speedup may vary based on workload","No documentation of memory bandwidth requirements or GPU memory utilization; exact VRAM requirements unknown"],"requires":["NVIDIA GPU with CUDA Compute Capability 7.0+ (Volta or newer)","vLLM with Megablocks CUDA kernels (specific version not documented)","CUDA 11.8+ and cuDNN 8.0+","Python 3.8+"],"input_types":["text tokens (up to 32,768 context window)"],"output_types":["generated text tokens with optimized inference throughput"],"categories":["automation-workflow","code-generation-editing"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__cap_6","uri":"capability://automation.workflow.cloud.deployment.via.skypilot","name":"cloud-deployment-via-skypilot","description":"Enables cloud deployment of Mixtral via SkyPilot framework, which abstracts cloud infrastructure provisioning and management. SkyPilot handles GPU selection, instance provisioning, and vLLM endpoint deployment across cloud providers, reducing operational complexity for teams deploying Mixtral at scale. The framework integrates with vLLM for optimized inference, enabling rapid deployment of Mixtral endpoints without manual infrastructure configuration.","intents":["Deploy Mixtral inference endpoints to cloud infrastructure without manual infrastructure management","Scale Mixtral inference across multiple cloud providers using a unified deployment framework","Reduce operational overhead for teams managing production language model deployments","Evaluate cloud deployment costs for Mixtral vs commercial LLM APIs"],"best_for":["Teams deploying Mixtral at scale across cloud infrastructure","Organizations seeking to reduce operational overhead for LLM deployment and management","Developers evaluating cloud deployment options for open-source models"],"limitations":["SkyPilot abstraction adds operational complexity; debugging infrastructure issues requires understanding both SkyPilot and underlying cloud provider","Cloud deployment costs depend on GPU availability and pricing; no cost guarantees or optimization beyond vLLM inference efficiency","SkyPilot support for specific cloud providers and GPU types not documented; compatibility matrix unknown","No built-in monitoring, logging, or observability; requires integration with external monitoring tools"],"requires":["SkyPilot framework installed and configured","Cloud provider credentials (AWS, GCP, Azure, etc.)","vLLM with Megablocks CUDA kernels","Python 3.8+"],"input_types":["SkyPilot deployment configuration and text prompts"],"output_types":["deployed vLLM inference endpoint accessible via HTTP API"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__cap_7","uri":"capability://safety.moderation.reduced.bias.and.fairness.evaluation","name":"reduced-bias-and-fairness-evaluation","description":"Demonstrates reduced bias compared to Llama 2 70B on BBQ (Bias Benchmark for QA) evaluation, indicating improved fairness across demographic groups. The model was evaluated on BBQ benchmark which measures bias in question-answering tasks across protected attributes (gender, race, religion, etc.). Additionally evaluated on BOLD (Bias in Open-Ended Language Generation) benchmark, showing more positive sentiment than Llama 2 but with similar variance, suggesting different bias characteristics rather than elimination of bias.","intents":["Deploy language models with documented lower bias compared to alternative open-source models","Evaluate fairness characteristics of language models using standard bias benchmarks","Understand bias-performance tradeoffs when selecting between open-source models","Build applications with fairness considerations using models with documented bias evaluation"],"best_for":["Teams building applications where fairness and bias reduction are critical requirements","Organizations evaluating open-source models for bias characteristics before deployment","Researchers studying bias in language models and fairness evaluation methodologies"],"limitations":["BBQ and BOLD benchmarks measure specific types of bias; performance on other fairness dimensions (e.g., occupational bias, ability bias) not documented","Bias evaluation is relative to Llama 2 70B; absolute bias levels not quantified, only comparative results provided","BOLD benchmark shows more positive sentiment than Llama 2; unclear if this represents reduced bias or different sentiment distribution","No documentation of bias mitigation techniques used during training or fine-tuning; bias reduction may be incidental rather than intentional"],"requires":["GPU with 24GB+ VRAM","vLLM or similar inference framework","Python 3.8+","BBQ and BOLD benchmark datasets for evaluation"],"input_types":["text prompts for bias evaluation (QA and open-ended generation tasks)"],"output_types":["generated text for bias measurement and fairness evaluation"],"categories":["safety-moderation","data-processing-analysis"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__cap_8","uri":"capability://tool.use.integration.apache.20.open.source.licensing","name":"apache-20-open-source-licensing","description":"Distributed under Apache 2.0 open-source license, enabling unrestricted commercial use, modification, and redistribution with minimal attribution requirements. The open weights are available for download and self-hosting, providing full control over model deployment, fine-tuning, and integration without reliance on proprietary APIs or vendor lock-in. Apache 2.0 licensing permits both commercial and non-commercial use cases with explicit patent protection.","intents":["Deploy Mixtral in commercial applications without licensing restrictions or API dependencies","Fine-tune and modify Mixtral for domain-specific tasks without vendor restrictions","Self-host Mixtral with full control over data, inference, and model behavior","Build products and services based on Mixtral without proprietary licensing costs"],"best_for":["Organizations requiring open-source models for commercial deployment without licensing restrictions","Teams building proprietary products that require model customization and control","Developers prioritizing vendor independence and avoiding API lock-in"],"limitations":["Apache 2.0 license requires attribution in derivative works; commercial products must include license notice","No warranty or liability protection; users assume all responsibility for model behavior and outputs","Open weights enable model redistribution; no mechanism to prevent unauthorized redistribution or commercial use by competitors","License compliance requires legal review for specific use cases; ambiguity around fine-tuned model licensing"],"requires":["Apache 2.0 license compliance documentation","Attribution notice in derivative works","No proprietary API keys or vendor accounts required"],"input_types":["model weights and configuration files"],"output_types":["fine-tuned models, derivative works, and commercial products"],"categories":["tool-use-integration","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__cap_9","uri":"capability://text.generation.language.32k.token.context.window","name":"32k-token-context-window","description":"Supports 32,768 token context window, enabling processing of long documents, multi-turn conversations, and complex prompts without chunking or context truncation. The extended context window is implemented through standard transformer architecture without explicit long-context techniques (e.g., ALiBi, RoPE modifications), relying on training data that includes long sequences. Context window enables coherent reasoning across longer documents and more complex multi-step tasks compared to models with smaller context windows.","intents":["Process long documents (up to ~24K tokens of text) without chunking or summarization","Maintain coherent multi-turn conversations with extensive history without context loss","Perform complex reasoning tasks requiring access to large amounts of context","Analyze and summarize long documents in a single inference pass"],"best_for":["Applications requiring long-context processing (document analysis, multi-turn conversation, complex reasoning)","Teams processing documents longer than typical context windows (4K-8K tokens)","Developers building applications where context truncation would degrade quality"],"limitations":["32K token context is hard limit; documents longer than this require chunking or summarization","Inference latency increases with context length; longer contexts require more computation and memory","Long-context performance not explicitly benchmarked; quality degradation at maximum context length unknown","Memory requirements scale with context length; 32K context requires significantly more VRAM than shorter contexts"],"requires":["GPU with sufficient VRAM for 32K token context (estimated 32GB+ for full precision)","vLLM or similar framework supporting long-context inference","Python 3.8+"],"input_types":["text documents and prompts up to 32,768 tokens"],"output_types":["generated text with access to full context window"],"categories":["text-generation-language","memory-knowledge"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"mixtral-8x7b__headline","uri":"capability://text.generation.language.sparse.mixture.of.experts.language.model","name":"sparse mixture-of-experts language model","description":"Mixtral 8x7B is a state-of-the-art sparse mixture-of-experts language model that delivers GPT-3.5-level performance while optimizing inference speed by activating only a fraction of its parameters, making it ideal for efficient text generation and instruction following.","intents":["best sparse mixture-of-experts model","sparse mixture-of-experts for text generation","fast language model for code generation","instruction-following model comparison","high-performance AI model for deployment"],"best_for":["efficient inference","high-performance text generation"],"limitations":["limited context window of 32,000 tokens"],"requires":[],"input_types":["text prompts"],"output_types":["text responses"],"categories":["text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":57,"verified":false,"data_access_risk":"high","permissions":["CUDA-capable GPU with sufficient VRAM (exact requirements not specified; estimated 24GB+ for full precision inference)","vLLM with Megablocks CUDA kernels for optimized sparse inference","Python 3.8+ for inference frameworks","GPU with sufficient VRAM for inference (estimated 24GB+ for full precision, 12GB+ for quantized variants)","vLLM or similar inference framework for efficient serving","Python 3.8+ for integration","Benchmark datasets (MMLU, HellaSwag, TruthfulQA, Winogrande, GSM8K, MATH, HumanEval, MT-Bench)","Evaluation framework (e.g., lm-evaluation-harness)","GPU with sufficient VRAM for inference","Python 3.8+"],"failure_modes":["Router network design and load-balancing strategy not publicly documented; potential for uneven expert utilization or routing instability under adversarial inputs","Sparse activation introduces routing overhead (~5-10% estimated) that is not quantified in official benchmarks","No documented failure modes for routing decisions or expert saturation scenarios","Routing decisions are per-token per-layer, creating potential for token-specific bottlenecks if certain experts become overloaded","Base model has no built-in safety guardrails; requires explicit prompting or preference tuning to refuse harmful requests","Context window hard-limited to 32,768 tokens; cannot process documents longer than this without chunking","Performance on specialized domains (medical, legal, scientific) not explicitly documented; general benchmarks may not reflect domain-specific capability gaps","Hallucination and factual accuracy not quantified; relies on same training data quality as other open-web-trained models","Benchmark scores reflect performance on specific evaluation datasets; real-world performance may differ","Benchmarks measure narrow capabilities (MMLU knowledge, HumanEval code); broader capability assessment requires additional evaluation","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-8x7b","compare_url":"https://unfragile.ai/compare?artifact=mixtral-8x7b"}},"signature":"xP8HsTSC5MDpzE9UrEDnux0zPzHRIC6JR2PoiNOAt+4IKhl9yWpV5U1Ih2cqX0xKdWYcaBgXjktwsH++PSw8Dg==","signedAt":"2026-06-20T22:37:55.789Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/mixtral-8x7b","artifact":"https://unfragile.ai/mixtral-8x7b","verify":"https://unfragile.ai/api/v1/verify?slug=mixtral-8x7b","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"}}