2763 - 面向企业用例的终极模型灵活性.pdf

《2763 - 面向企业用例的终极模型灵活性.pdf》由会员分享，可在线阅读，更多相关《2763 - 面向企业用例的终极模型灵活性.pdf（11页珍藏版）》请在三个皮匠报告上搜索。

1、WatsonWatsonx x.ai.aiUltimate Model Flexibility for Enterprise Use CasesLast update:September 29th,2025Model Gateway2IBM watsonx.aiModel As a ServiceDeploy on Demand CatalogBring your own custom models-Central AI Control Plane to Central AI Control Plane to access any model access any model anywhere

2、3 3rdrd Party Hosted ModelsParty Hosted ModelsAccess SOTA non-hosted 3rd-party models through a secure routerOverviewMulti-tenant SaaS offering where capacity is sharedSingle-tenant SaaS offering with curated set of models from IBMSingle-tenant SaaS offering with customer-provided modelsFlexible ten

3、ancy depending on connected providerCost StructureToken-based pricing,billed on both input and output tokensHourly pricing,billed as long as model is deployedHourly pricing,billed as long as model is deployedPricing dependent on connected model provider and modelBenefitsPay only for the tokens that

4、are consumed Indemnity offered for IBM and select modelsPredictable cost based on model deploymentDedicated GPU capacityEase of deployment No rate limitsIndemnity offered for IBM and select models Predictable cost based on model deploymentDedicated GPU capacityNo rate limitsUsing OpenAI API compatib

5、le endpointsAbility to manage model accessLargest model availability(incl.watsonx.ai hosted models)DrawbacksGPU capacity is shared;unable to take full advantage of the throughputDeprovision the model when not in use to optimize costsDeprovision the model when not in use to optimize costsRequires use

6、r to import in relevant model filesData leaves watsonx.ai servers(opt-in)Possible additional latency from calling out to third-party hosted modelsBest Fit forExperimentationSmall companies with infrequent useLarge production workloads Enterprises with SLA requirementsLarge production workloads Enter