{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"ai21-jamba-1-5","slug":"ai21-jamba-1-5","name":"AI21 Jamba 1.5","type":"model","url":"https://www.ai21.com/jamba","page_url":"https://unfragile.ai/ai21-jamba-1-5","categories":["model-training","documentation"],"tags":[],"pricing":{"model":"free","free":true,"starting_price":null},"status":"active","verified":false},"capabilities":[{"id":"ai21-jamba-1-5__cap_0","uri":"capability://text.generation.language.hybrid.mamba.transformer.long.context.text.generation","name":"hybrid mamba-transformer long-context text generation","description":"Generates text using a hybrid architecture that interleaves Mamba structured state space (SSS) layers with Transformer attention layers, enabling linear-time sequence processing instead of quadratic complexity. The Mamba layers maintain recurrent state across 256K token contexts while Transformer layers provide attention-based refinement, allowing efficient inference on documents up to 256K tokens without the memory explosion of pure Transformer models. This architecture enables processing of entire books, legal contracts, or multi-document datasets in a single forward pass.","intents":["Process entire long documents (financial reports, legal contracts, research papers) without chunking or context windowing","Perform multi-document reasoning and synthesis across dozens of related documents simultaneously","Build RAG systems that can ingest full documents without truncation or lossy summarization","Generate coherent responses that maintain consistency across very long input contexts"],"best_for":["Enterprise teams processing financial documents, legal contracts, and regulatory filings","Researchers and analysts working with multi-document datasets requiring holistic understanding","RAG system builders needing to preserve full document context without chunking","Organizations with memory-constrained infrastructure seeking efficient long-context inference"],"limitations":["Hard context window limit of 256K tokens (~200K words); documents exceeding this require truncation or multi-pass processing","Mamba layers use recurrent state which may degrade performance on tasks requiring precise attention to distant context (unknown degradation curve at max context)","No quantitative benchmarks provided comparing long-context performance to GPT-4 Turbo or Claude 3.5 Sonnet on standard long-context tasks","Fine-tuning methodology for long-context tasks not documented; unclear if standard instruction-tuning preserves long-context capabilities"],"requires":["API access via AI21 Studio (free $10 trial available) or self-hosted deployment","For self-hosting: GPU with sufficient VRAM (exact requirements unknown; claims 'significantly less memory' than comparable models but no absolute specs provided)","Text input in supported format (plain text, markdown, or structured documents)"],"input_types":["text (plain text, markdown, HTML, PDF-as-text)","multi-document collections (up to 256K tokens combined)"],"output_types":["text (generated continuations, summaries, analyses, answers)"],"categories":["text-generation-language","long-context-processing"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"ai21-jamba-1-5__cap_1","uri":"capability://text.generation.language.instruction.following.chat.with.enterprise.domain.knowledge","name":"instruction-following chat with enterprise domain knowledge","description":"Provides instruction-following and conversational capabilities through fine-tuned Chat and Instruct variants optimized for enterprise use cases across Finance, Tech, Defense, Healthcare, and Manufacturing domains. The model follows natural language instructions with context awareness maintained across the 256K token window, enabling multi-turn conversations that reference earlier context without degradation. Deployed via AI21 Studio API with usage-based pricing or self-hosted on customer infrastructure.","intents":["Build enterprise chatbots that understand domain-specific terminology and context across long conversations","Create instruction-following agents that can process complex, multi-step requests with reference to historical context","Deploy conversational interfaces for customer support, research assistance, or internal knowledge workers","Fine-tune or prompt-engineer the model for domain-specific instruction following without retraining"],"best_for":["Enterprise teams in Finance, Healthcare, Defense, or Manufacturing needing domain-aware conversational AI","Organizations building internal knowledge worker assistants that must reference long conversation histories","Teams deploying chatbots where context retention across 50+ turn conversations is critical","Builders requiring instruction-following without the latency of larger models (Jamba Mini: ~12B active parameters)"],"limitations":["Fine-tuning methodology not documented; unclear if custom instruction-tuning is supported or only prompt-based customization","No domain-specific pre-trained variants provided; enterprises must implement their own domain adaptation via prompting or fine-tuning","Benchmark performance on instruction-following tasks not quantified; only qualitative claims of 'outperforming comparable models'","No explicit safety guidelines or content filtering documentation; moderation approach unknown"],"requires":["API key for AI21 Studio ($0.2/1M input tokens for Mini, $2/1M for Large) or self-hosted deployment","For self-hosting: Python 3.9+ and appropriate GPU/CPU hardware (specs unknown)","Structured prompt engineering or fine-tuning data if customizing for domain-specific behavior"],"input_types":["text (natural language instructions, multi-turn conversation history up to 256K tokens)"],"output_types":["text (instruction responses, conversational replies, domain-specific answers)"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"ai21-jamba-1-5__cap_10","uri":"capability://tool.use.integration.open.source.model.weights.and.community.deployment","name":"open-source model weights and community deployment","description":"Jamba models are released as open-source with weights available on Hugging Face, enabling community contributions, research, and custom deployments. The open-source approach allows researchers to study the hybrid Mamba-Transformer architecture, contribute improvements, and build upon the models. Community members can create optimized inference implementations, fine-tuning guides, and domain-specific adaptations without licensing restrictions.","intents":["Research and study the hybrid Mamba-Transformer architecture and its effectiveness","Contribute improvements and optimizations to the model architecture or inference","Build custom applications and integrations without licensing or commercial restrictions","Create community-driven fine-tuning guides and domain-specific adaptations"],"best_for":["Researchers studying efficient language model architectures and state space models","Open-source contributors and community builders","Organizations with strong open-source cultures and community engagement","Academic institutions and non-profit organizations"],"limitations":["License terms not specified in provided materials; unclear if models are under Apache 2.0, MIT, or other open-source license","No official community governance or contribution guidelines documented","Community support and documentation quality depend on community engagement; may be limited compared to well-funded projects","No guarantee of long-term maintenance or updates from AI21 Labs","Commercial use restrictions (if any) not documented"],"requires":["Hugging Face account and familiarity with open-source model repositories","Understanding of open-source licensing and usage rights","Community engagement and contribution guidelines (if contributing)"],"input_types":["model weights and architecture code"],"output_types":["community contributions, optimizations, and adaptations"],"categories":["tool-use-integration","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"ai21-jamba-1-5__cap_2","uri":"capability://text.generation.language.efficient.inference.with.reduced.memory.footprint","name":"efficient inference with reduced memory footprint","description":"Achieves inference efficiency through the Mamba SSS architecture which eliminates the quadratic memory scaling of Transformer self-attention, reducing GPU VRAM requirements compared to models of similar capability. The hybrid design balances efficiency gains from Mamba layers with quality preservation from Transformer layers, enabling deployment on resource-constrained infrastructure. Supports both API-based inference via AI21 Studio and self-hosted deployment with configurable hardware.","intents":["Deploy large language models on edge devices, laptops, or cost-constrained cloud infrastructure","Reduce inference latency and VRAM requirements for real-time applications like chatbots or content generation","Run long-context inference (256K tokens) on hardware that would require prohibitive VRAM for pure Transformer models","Optimize inference cost by reducing GPU requirements while maintaining competitive model quality"],"best_for":["Teams with GPU-constrained infrastructure (limited VRAM, edge deployment, cost-sensitive cloud budgets)","Builders of real-time inference systems where latency is critical (sub-second response times)","Organizations seeking to minimize inference costs through efficient model architecture","Developers deploying on consumer-grade GPUs or CPU-only environments"],"limitations":["Exact GPU VRAM requirements unknown; documentation claims 'significantly less memory' than comparable models but provides no absolute specifications or comparison baselines","Inference speed benchmarks not provided; claims of 'fastest processing on the market' and 'remarkable processing speeds' are qualitative without latency metrics (tokens/sec, ms/token)","No quantitative comparison of inference efficiency vs. quantized versions of GPT-3.5, Llama 2, or other efficient models","Hardware requirements for self-hosting not documented; unclear if deployment requires GPU or if CPU inference is viable"],"requires":["For API inference: AI21 Studio account with active credits ($0.2-$2/1M input tokens depending on variant)","For self-hosted: GPU with unknown VRAM requirement (claims efficiency but no specs) or CPU with sufficient RAM","Python 3.9+ for local deployment; specific framework dependencies unknown"],"input_types":["text (any length up to 256K tokens)"],"output_types":["text (generated output)"],"categories":["text-generation-language","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"ai21-jamba-1-5__cap_3","uri":"capability://tool.use.integration.api.based.inference.with.usage.based.pricing","name":"api-based inference with usage-based pricing","description":"Provides hosted inference via AI21 Studio API with transparent usage-based pricing ($0.2-$0.4/1M tokens for Mini, $2-$8/1M tokens for Large) and free trial credits ($10 for 3 months, no credit card required). Supports both Jamba Mini (12B active) and Large (94B active) variants with identical API interface, enabling cost-optimization by selecting appropriate model size per use case. Integrates with standard HTTP/REST patterns and SDKs for Python and other languages.","intents":["Prototype and deploy LLM applications without managing infrastructure or GPU hardware","Optimize inference costs by selecting between Mini (faster, cheaper) and Large (higher quality) variants per request","Access long-context inference (256K tokens) without provisioning expensive GPU infrastructure","Integrate Jamba into existing applications via standard REST APIs and language-specific SDKs"],"best_for":["Startups and small teams without dedicated ML infrastructure","Builders prototyping LLM applications and wanting to defer infrastructure decisions","Organizations with variable inference load seeking pay-as-you-go pricing without upfront commitment","Teams evaluating Jamba before committing to self-hosted deployment"],"limitations":["API endpoint specifications, rate limits, and request/response formats not documented in provided materials","Pricing is per-token with no volume discounts or reserved capacity options mentioned; cost scales linearly with usage","Free trial limited to $10 credits over 3 months (~50M tokens for Mini input); insufficient for production evaluation","No SLA, uptime guarantees, or latency SLOs documented; production reliability unknown","API availability and regional deployment not specified; unclear if multi-region or single-region service"],"requires":["AI21 Studio account (free signup, no credit card required for trial)","$10 trial credits or active payment method for production use","API key for authentication","Python SDK or HTTP client library for integration"],"input_types":["text (up to 256K tokens per request)"],"output_types":["text (generated output with token counts for billing)"],"categories":["tool-use-integration","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"ai21-jamba-1-5__cap_4","uri":"capability://automation.workflow.self.hosted.deployment.with.private.infrastructure","name":"self-hosted deployment with private infrastructure","description":"Enables deployment of Jamba models on customer-controlled infrastructure (on-premises or private cloud) via model downloads from Hugging Face and integration with standard inference frameworks. Supports deployment through 'trusted technology partners' (partners not named in documentation) and custom cloud deployments. Provides full model control, data privacy, and elimination of API latency at the cost of infrastructure management and operational complexity.","intents":["Deploy Jamba with full data privacy and control, keeping all inputs/outputs on customer infrastructure","Integrate Jamba into existing ML infrastructure and deployment pipelines","Optimize inference latency by eliminating API round-trip overhead for real-time applications","Customize model behavior through fine-tuning or quantization without vendor lock-in"],"best_for":["Enterprises with strict data privacy requirements (financial services, healthcare, defense)","Organizations with existing ML infrastructure and DevOps capabilities","Teams building latency-sensitive applications where API overhead is unacceptable","Builders planning long-term deployment and willing to manage infrastructure"],"limitations":["Hardware requirements not documented; no guidance on minimum/recommended GPU VRAM, CPU, or memory for self-hosting either variant","Model format and quantization options unknown; unclear if available as safetensors, PyTorch, GGUF, or other formats","Inference framework compatibility not specified; unclear if compatible with vLLM, TensorRT, ONNX, or other standard frameworks","Fine-tuning methodology and tooling not documented; unclear if standard Hugging Face fine-tuning works or if custom tooling required","No deployment guides, Docker images, or reference architectures provided in available documentation","Operational support and SLAs for self-hosted deployments unknown"],"requires":["GPU with sufficient VRAM (exact requirement unknown; claims efficiency but no specs provided)","Python 3.9+ and standard ML frameworks (PyTorch, Transformers library, etc.)","Hugging Face account to download model weights","Infrastructure for hosting (on-premises servers, cloud VMs, or Kubernetes cluster)","DevOps/ML engineering expertise to manage deployment, scaling, and monitoring"],"input_types":["text (up to 256K tokens)"],"output_types":["text (generated output)"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"ai21-jamba-1-5__cap_5","uri":"capability://text.generation.language.multi.document.synthesis.and.comparison","name":"multi-document synthesis and comparison","description":"Leverages the 256K context window to simultaneously process and synthesize information across multiple related documents (financial reports, research papers, contracts, etc.) in a single inference pass. The hybrid Mamba-Transformer architecture maintains coherent understanding across document boundaries while the linear-time complexity enables processing of dozens of documents without memory explosion. Enables cross-document reasoning, contradiction detection, and synthesis without lossy summarization or chunking.","intents":["Compare financial statements across multiple quarters or competitors to identify trends and anomalies","Synthesize findings across dozens of research papers or technical documents to identify consensus and conflicts","Analyze contract terms across multiple agreements to identify inconsistencies or compliance risks","Build knowledge base search systems that return synthesized answers across multiple source documents"],"best_for":["Financial analysts and compliance teams comparing multiple documents","Researchers synthesizing findings across large literature reviews","Legal teams analyzing contract portfolios for consistency and risk","Knowledge management teams building enterprise search and synthesis systems"],"limitations":["Hard limit of 256K tokens (~200K words) means large document collections must be curated or truncated","No documented methodology for handling contradictions or conflicting information across documents","Synthesis quality not benchmarked; unclear how performance compares to human analysts or multi-pass approaches","No built-in document attribution or source tracking; unclear if model can reliably cite which document each claim comes from","Fine-tuning for domain-specific synthesis not documented; unclear if standard instruction-tuning preserves multi-document reasoning"],"requires":["API access via AI21 Studio or self-hosted deployment","Documents pre-processed and concatenated within 256K token limit","Structured prompts that clearly delineate document boundaries and synthesis objectives"],"input_types":["text (multiple documents concatenated, total up to 256K tokens)"],"output_types":["text (synthesized analysis, comparisons, cross-document insights)"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"ai21-jamba-1-5__cap_6","uri":"capability://text.generation.language.efficient.tokenization.with.30.compression","name":"efficient tokenization with 30% compression","description":"Claims to achieve up to 30% more text per token than competing providers through optimized tokenization, reducing the effective cost of long-context processing and enabling more content to fit within the 256K token window. The tokenization approach is not documented, but the claim suggests more efficient encoding of natural language compared to standard BPE or SentencePiece tokenizers used by other models.","intents":["Reduce effective API costs by fitting more text into the same token budget","Process longer documents within the 256K token limit without truncation","Optimize token usage for cost-sensitive applications with high volume"],"best_for":["Cost-sensitive applications processing large volumes of text","Teams optimizing token budgets for long-context inference","Organizations comparing per-token costs across providers"],"limitations":["Tokenization methodology not documented; no explanation of how 30% compression is achieved","No verification or independent benchmarking of the 30% claim; methodology unknown","Compression may vary by language, domain, or text type; no breakdown provided","Unclear if compression applies equally to input and output tokens or only input"],"requires":["API access via AI21 Studio or self-hosted deployment"],"input_types":["text"],"output_types":["text"],"categories":["text-generation-language","data-processing-analysis"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"ai21-jamba-1-5__cap_7","uri":"capability://tool.use.integration.open.source.model.weights.with.hugging.face.distribution","name":"open-source model weights with hugging face distribution","description":"Distributes Jamba model weights via Hugging Face Model Hub as open-source models, enabling free download, inspection, and modification without licensing restrictions. Both Mini (12B active/52B total) and Large (94B active/398B total) variants are available, allowing developers to use, fine-tune, and redistribute models under open-source terms. Supports integration with standard Hugging Face tooling (transformers library, model cards, community discussions).","intents":["Download and inspect model weights for research, auditing, or understanding architecture","Fine-tune Jamba on custom datasets using standard Hugging Face fine-tuning tools","Build derivative models or research variants without licensing restrictions","Integrate Jamba into open-source projects and frameworks"],"best_for":["Researchers and academics studying long-context architectures","Open-source projects and communities","Organizations with open-source-first policies","Teams planning to fine-tune or modify models"],"limitations":["License type not explicitly stated in documentation; unclear if Apache 2.0, MIT, or other open-source license","Commercial use restrictions unknown; unclear if commercial deployment is permitted under license terms","Model card and documentation quality unknown; may lack detailed information on training data, biases, or limitations","Community support and maintenance unclear; unknown if AI21 Labs actively maintains models or community-driven"],"requires":["Hugging Face account (free)","Python 3.9+ and transformers library","GPU or CPU with sufficient resources to download and run models"],"input_types":["text"],"output_types":["text"],"categories":["tool-use-integration","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"ai21-jamba-1-5__cap_8","uri":"capability://text.generation.language.parameter.efficient.inference.with.mixture.of.experts.style.sparsity","name":"parameter-efficient inference with mixture-of-experts-style sparsity","description":"Jamba Mini uses only 12B active parameters out of 52B total parameters through sparse activation patterns, and Jamba Large uses 94B active per 398B total, enabling inference with reduced computational cost compared to dense models of equivalent quality. The hybrid architecture with Mamba layers contributes to this efficiency by avoiding the dense attention computations of pure Transformers. This sparsity pattern is similar to mixture-of-experts approaches but implemented through the Mamba-Transformer hybrid design.","intents":["Achieve model quality comparable to larger dense models while using fewer active parameters","Reduce inference latency and computational cost through sparse activation","Deploy larger effective model capacity without proportional increases in inference cost"],"best_for":["Teams seeking to balance model quality with inference efficiency","Cost-sensitive deployments where reducing active parameters is critical","Real-time inference systems where latency is constrained"],"limitations":["Sparsity mechanism not documented; unclear how 12B active vs. 52B total is achieved (mixture-of-experts, pruning, conditional computation, etc.)","No benchmarking of quality vs. dense models of equivalent active parameter count","Inference framework support for sparse activation unknown; unclear if standard frameworks can exploit sparsity"],"requires":["API access or self-hosted deployment"],"input_types":["text"],"output_types":["text"],"categories":["text-generation-language","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"ai21-jamba-1-5__cap_9","uri":"capability://text.generation.language.enterprise.domain.specific.deployment","name":"enterprise domain-specific deployment","description":"Positions Jamba for enterprise use across Finance, Tech, Defense, Healthcare, and Manufacturing domains with claims of domain-specific optimization, though no domain-specific model variants or fine-tuning details are documented. The 256K context window and efficient inference enable deployment in enterprise environments with large document volumes and strict latency/privacy requirements. Available through 'trusted technology partners' for cloud deployment (partners not named).","intents":["Deploy domain-aware LLM in regulated industries (Finance, Healthcare, Defense) with data privacy and compliance","Process industry-specific documents (financial statements, medical records, technical specifications) with domain understanding","Integrate Jamba into enterprise applications with existing infrastructure and compliance frameworks"],"best_for":["Financial services firms analyzing documents and regulatory filings","Healthcare organizations processing medical records and research","Defense and government agencies with strict data privacy requirements","Manufacturing companies analyzing technical documentation and specifications"],"limitations":["No domain-specific model variants provided; enterprises must implement domain adaptation via prompting or fine-tuning","No documentation of domain-specific fine-tuning or evaluation; unclear if models are optimized for industry terminology or compliance","Trusted technology partners not named; unclear which cloud providers or integrators support Jamba","Compliance certifications (SOC 2, HIPAA, FedRAMP, etc.) not documented; unclear if Jamba meets regulatory requirements","No industry-specific benchmarks or case studies provided"],"requires":["API access via AI21 Studio or deployment through named partner (partners unknown)","For self-hosted: infrastructure meeting enterprise compliance requirements"],"input_types":["text (domain-specific documents)"],"output_types":["text (domain-specific analysis and responses)"],"categories":["text-generation-language","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"ai21-jamba-1-5__headline","uri":"capability://text.generation.language.long.context.language.model.for.document.understanding","name":"long-context language model for document understanding","description":"AI21 Jamba 1.5 is a cutting-edge language model designed for long document understanding and multi-document tasks, featuring a massive 256K context window and efficient inference.","intents":["best long-context language model","language model for document understanding","AI model for multi-document tasks","top AI model for long-context benchmarks","best model for efficient long document processing"],"best_for":["long document analysis","multi-document processing"],"limitations":["maximum context window of 256,000 tokens"],"requires":[],"input_types":["text"],"output_types":["text"],"categories":["text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":59,"verified":false,"data_access_risk":"high","permissions":["API access via AI21 Studio (free $10 trial available) or self-hosted deployment","For self-hosting: GPU with sufficient VRAM (exact requirements unknown; claims 'significantly less memory' than comparable models but no absolute specs provided)","Text input in supported format (plain text, markdown, or structured documents)","API key for AI21 Studio ($0.2/1M input tokens for Mini, $2/1M for Large) or self-hosted deployment","For self-hosting: Python 3.9+ and appropriate GPU/CPU hardware (specs unknown)","Structured prompt engineering or fine-tuning data if customizing for domain-specific behavior","Hugging Face account and familiarity with open-source model repositories","Understanding of open-source licensing and usage rights","Community engagement and contribution guidelines (if contributing)","For API inference: AI21 Studio account with active credits ($0.2-$2/1M input tokens depending on variant)"],"failure_modes":["Hard context window limit of 256K tokens (~200K words); documents exceeding this require truncation or multi-pass processing","Mamba layers use recurrent state which may degrade performance on tasks requiring precise attention to distant context (unknown degradation curve at max context)","No quantitative benchmarks provided comparing long-context performance to GPT-4 Turbo or Claude 3.5 Sonnet on standard long-context tasks","Fine-tuning methodology for long-context tasks not documented; unclear if standard instruction-tuning preserves long-context capabilities","Fine-tuning methodology not documented; unclear if custom instruction-tuning is supported or only prompt-based customization","No domain-specific pre-trained variants provided; enterprises must implement their own domain adaptation via prompting or fine-tuning","Benchmark performance on instruction-following tasks not quantified; only qualitative claims of 'outperforming comparable models'","No explicit safety guidelines or content filtering documentation; moderation approach unknown","License terms not specified in provided materials; unclear if models are under Apache 2.0, MIT, or other open-source license","No official community governance or contribution guidelines documented","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.7,"quality":0.9,"ecosystem":0.39999999999999997,"match_graph":0.25,"freshness":0.9,"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":"2026-05-24T12:16:19.836Z","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=ai21-jamba-1-5","compare_url":"https://unfragile.ai/compare?artifact=ai21-jamba-1-5"}},"signature":"XpWTPYRYka+f1t/ifskOmfHhOoyhQbLZr9v71TzQ35VouoIdcBVod9Yh5euDiWbRz0xMn6nueRdbeUxqFN4eCQ==","signedAt":"2026-06-15T05:34:05.444Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/ai21-jamba-1-5","artifact":"https://unfragile.ai/ai21-jamba-1-5","verify":"https://unfragile.ai/api/v1/verify?slug=ai21-jamba-1-5","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"}}