# Zyphra Access hosted AI model inference with serverless service and dedicated capacity options. - Findings24: - Site: - Categorias: Plataformas de inferência de IA ## O que é Zyphra? Zyphra combines an open AI research lab with a cloud platform for building and deploying AI systems. The current Zyphra Cloud offering includes hosted model serving through Zyphra Inference, with a focus on long-context and multi-step agent workloads. Developers can access the cloud service rather than operate the model-serving infrastructure themselves. The platform offers serverless inference and dedicated inference capacity. Its model catalog includes language models and speech generation, with usage prices that depend on the selected model and input type. Zyphra Research investigates architectures and computing systems, but those research topics are distinct from the capabilities of a hosted service. Research topics and announced platform plans do not establish that the corresponding capabilities are currently available in Cloud. ## O que Zyphra pode fazer? ### Serverless inference Run hosted open-weight models without managing the serving infrastructure. Access the service through Zyphra Cloud and its API offering. ### Dedicated inference capacity Use reserved serving capacity for latency-sensitive production workloads rather than a general serverless deployment. ### Model-specific serving and prices The Cloud model catalog determines which hosted language and speech capabilities are available. Check each model’s current availability, usage units, and rates separately from research announcements. ## Casos de uso ### Serve an agent application Use hosted inference for a multi-step application where long context and model-serving infrastructure are important requirements. ### Compare inference deployment options Evaluate serverless access against reserved capacity for an application with specific latency and capacity needs. ### Add language or speech generation Select the appropriate hosted model for language tasks or text-to-speech, and account for its own usage unit and price. ## Perguntas frequentes ### How are model usage prices measured? Usage units and prices depend on the selected model and service. Check the current Zyphra Cloud pricing before use, including how input, output, and any cached usage are measured. Do not assume that tokens, bytes, characters, and audio minutes are interchangeable. ### Is any hosted inference listed at no charge? Available models, any no-charge options, and usage allowances depend on the current Cloud catalog and pricing. An open research model or its license does not by itself establish free hosted inference or free reserved capacity. ### Does research in long-term memory mean the cloud supplies it as a finished feature? No. Research directions and announced platform plans should be distinguished from services currently available in Zyphra Cloud. Confirm the availability of a specific capability in the current service documentation before planning a deployment. ### Can I use a public demo for a production workload? A public demo should not be treated as a production service commitment. Review the applicable service and data-handling terms before production use, and confirm the handling conditions before submitting confidential or sensitive information. ## Informações rápidas - **Core hosted offering**: Zyphra Cloud model inference - **Deployment options**: Serverless inference and dedicated serving capacity - **Primary users**: Developers and enterprises integrating AI model services - **Usage units**: Current billing units for the selected model and service - **Related work**: Open-model and computing-systems research