> ## Documentation Index > Fetch the complete documentation index at: https://fireworks.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Get Account Usage ## OpenAPI ````yaml get /v1/accounts/{account_id}/billingUsage openapi: 3.1.0 info: title: Gateway REST API version: 5.10.0 servers: - url: https://api.fireworks.ai security: - BearerAuth: [] tags: - name: AccountService - name: DeploymentService - name: Gateway - name: ModelService - name: TrainingService paths: /v1/accounts/{account_id}/billingUsage: get: tags: - Gateway summary: Get Account Usage operationId: Gateway_GetAccountUsage parameters: - name: startTime description: |- Costs returned are inclusive of `start_time`. start_time must be before end_time. in: query required: true schema: type: string format: date-time - name: endTime description: |- Costs returned are exclusive of `end_time`. end_time must not be more than 31 days after start_time. in: query required: true schema: type: string format: date-time - name: usageType description: >- Usage type to query usage for If not specified, returns all usage types (serverless, dedicated deployments, and training). - USAGE_TYPE_UNSPECIFIED: Default value. When specified (or when usage_type field is not set), returns usage data for all deployment types: serverless requests, dedicated deployments, and training jobs. - SERVERLESS: Returns only serverless usage data. Filters the response to include only usage from serverless API requests. - DEDICATED_DEPLOYMENT: Returns only dedicated deployment usage data. Filters the response to include only usage from dedicated deployments. - TRAINING: Returns only training job usage data (SFT/DPO token usage and RFT / service-mode trainer GPU-seconds usage). Inference deployments serving rollouts for RFT / online RL are reported under DEDICATED_DEPLOYMENT (not TRAINING) to avoid double counting GPU time. in: query required: false schema: type: string enum: - USAGE_TYPE_UNSPECIFIED - SERVERLESS - DEDICATED_DEPLOYMENT - TRAINING default: USAGE_TYPE_UNSPECIFIED - name: timezone description: >- IANA timezone identifier for daily aggregation (e.g., "America/Los_Angeles", "Europe/London"). When specified, the returned data will be aggregated into daily buckets based on this timezone. If not specified or empty, defaults to "UTC". See: https://en.wikipedia.org/wiki/List_of_tz_database_time_zones in: query required: false schema: type: string - name: groupBy description: >- Dimensions to group usage by (multiple values allowed; each is a separate GROUP BY column). Serverless: "model_name", "api_key_id", "api_key_name", "user_id", "session_id", "annotations.team", "annotations.project", "annotations.environment". Dedicated: "deployment_name", "accelerator_type", and the same annotation keys. Training: "job_id", "training_session_id", "job_type", "usage_type", "accelerator_type", "base_model", and the same annotation keys. When usage_type is unspecified, dimensions that apply only to one stream are ignored on the others (e.g. "deployment_name" is ignored for serverless and training; "model_name" / "api_key_id" / "api_key_name" / "user_id" / "session_id" are ignored for dedicated and training; "job_id" / "training_session_id" / "job_type" are ignored for serverless and dedicated). "session_id" is the opaque client session from billing properties (serverless-only, high cardinality): prefer filtering by a single session; grouping by it can fan out to one row per session. Example: ["annotations.team", "model_name"] or ["user_id", "model_name"] or ["api_key_id", "api_key_name"]. If empty: serverless aggregates by model name; dedicated defaults to deployment and accelerator type; training aggregates by job_id, training_session_id, job_type, usage_type, accelerator_type and base_model. in: query required: false explode: true schema: type: array items: type: string - name: filter description: >- This is a request variable of the map type. The query format is "map_name[key]=value", e.g. If the map name is Age, the key type is string, and the value type is integer, the query parameter is expressed as Age["bob"]=18 in: query required: false schema: type: string - name: account_id in: path required: true description: The Account Id schema: type: string responses: '200': description: A successful response. content: application/json: schema: $ref: '#/components/schemas/gatewayAccountUsage' components: schemas: gatewayAccountUsage: type: object properties: serverlessCosts: type: array items: $ref: '#/components/schemas/AccountUsageServerlessUsage' type: object title: List of serverless cost data dedicatedCosts: type: array items: $ref: '#/components/schemas/AccountUsageDedicatedDeploymentUsage' type: object title: List of dedicated deployment cost data trainingCosts: type: array items: $ref: '#/components/schemas/AccountUsageTrainingUsage' type: object description: >- List of training costs (SFT / DPO token usage and RFT / service-mode accelerator-second usage). Empty when training usage was not requested or when the account has training usage hidden. title: Response with model costs by deployment type AccountUsageServerlessUsage: type: object properties: modelName: type: string description: 'Deprecated: use group["model_name"] instead.' promptTokens: type: string format: int64 title: Number of prompt tokens (for text inference) completionTokens: type: string format: int64 title: Number of completion tokens (for text inference) cachedPromptTokens: type: string format: int64 description: >- Number of prompt tokens served from cache (for text inference). Subset of prompt_tokens. uncachedPromptTokens: type: string format: int64 description: >- Number of prompt tokens not served from cache (for text inference). prompt_tokens - cached_prompt_tokens. startTime: type: string format: date-time title: >- Start timestamp of the usage. Note: Serverless usage events are aggregated daily. Only the date portion (YYYY-MM-DD) is used; endTime: type: string format: date-time title: >- End timestamp of the usage. Note: Serverless usage events are aggregated daily. Only the date portion (YYYY-MM-DD) is used; audioInputSeconds: type: number format: double title: Audio input seconds (for audio inference) usageType: type: string title: Usage type to distinguish between different inference types apiKeyId: type: string description: 'Deprecated: use group["api_key_id"] instead.' group: type: object additionalProperties: type: string description: >- Group-by dimension key-values (model_name, api_key_id, api_key_name, user_id, session_id, team, project, environment). session_id is present only for buckets that had an opaque client session; no-session usage omits the key (all such usage collapses into a single "no session" bucket). costNanoUsd: type: number format: double description: >- Cost in nano-USD (1e-9 USD), summed from properties.cost. 0 when absent (not "free"). Only huggingface currently stamps authoritative cost. description: One serverless usage aggregation bucket. AccountUsageDedicatedDeploymentUsage: type: object properties: deploymentId: type: string title: The deployment ID acceleratorType: type: string title: GPU type / accelerator type acceleratorSeconds: type: string format: int64 title: Accelerator seconds startTime: type: string format: date-time description: >- Start timestamp of the usage. Note: Dedicated deployment usage events are aggregated daily. Only the date portion (YYYY-MM-DD) is used; the time portion is ignored. endTime: type: string format: date-time description: >- End timestamp of the usage. Note: Dedicated deployment usage events are aggregated daily. Only the date portion (YYYY-MM-DD) is used; the time portion is ignored. baseModel: type: string title: Base model of the deployment usageType: type: string title: Usage type to distinguish between different deployment types placement: type: string description: >- Deployment placement as a string (e.g. US_IOWA_1, GLOBAL, US, EUROPE). Set to REGION_UNSPECIFIED when unknown. group: type: object additionalProperties: type: string description: >- Group-by dimension key-values (deployment_name, accelerator_type, team, project, environment). description: One dedicated deployment usage aggregation bucket. AccountUsageTrainingUsage: type: object properties: jobId: type: string description: >- The managed training job id (e.g. supervised_fine_tuning_job id, rlor_trainer_job id). Empty for serverless training token rows; use `training_session_id` for those. Best-effort: extracted from billing event_id for legacy records that did not emit it as an explicit property. trainingSessionId: type: string description: >- The serverless training session id. Populated for SERVERLESS_TRAINING_TOKENS rows and empty for managed training jobs. jobType: type: string description: |- High-level training job kind: "sft", "dpo", "rft", "trainer", etc. Best-effort for legacy records. usageType: type: string description: >- The underlying billing usage_type as emitted to the billing pipeline, e.g. "SUPERVISED_FINE_TUNING_USAGE", "REINFORCEMENT_FINE_TUNING_GPU_SECONDS", "FULL_PARAM_FINE_TUNING_GPU_HOURS". Frontends can use this to pick the right metric to display (tokens vs. accelerator_seconds). tokens: type: string format: int64 description: |- Total tokens billed for this bucket. Populated for SFT / DPO usage; 0 for accelerator-second usage types. acceleratorType: type: string description: >- Accelerator type for accelerator-second usage (e.g. "NVIDIA_H100_80GB"). Empty for token-based usage. acceleratorSeconds: type: string format: int64 description: >- Total accelerator seconds billed for this bucket. Frontends typically display this as GPU hours (seconds / 3600). 0 for token-based usage. baseModel: type: string description: Base model for the training job, when known. startTime: type: string format: date-time description: >- Start timestamp of the usage bucket. Note: Training usage events are aggregated daily. Only the date portion (YYYY-MM-DD) is used; the time portion is ignored. endTime: type: string format: date-time description: >- End timestamp of the usage bucket. Note: Training usage events are aggregated daily. Only the date portion (YYYY-MM-DD) is used; the time portion is ignored. group: type: object additionalProperties: type: string description: >- Group-by dimension key-values for the requested group_by. May include any of: job_id, training_session_id, job_type, usage_type, accelerator_type, base_model, annotations.team, annotations.project, annotations.environment. Empty values are omitted. Annotation values will typically be empty for training usage until billing emitters propagate annotations. annotations: type: object additionalProperties: type: string description: |- Annotation key-values associated with this bucket (team, project, environment). Mirrors the annotation entries in `group` for API symmetry with future per-stream consumers. Typically empty for training usage today until trainer-side billing emitters populate properties.annotation_* on every record. description: >- One training usage aggregation bucket. Training usage is emitted by the trainer-side billing pipeline. SFT / DPO usage is metered in tokens; RFT and full-param (service-mode) trainer usage is metered in accelerator-seconds. Both metrics are returned on the same message; consumers should use `usage_type` to disambiguate which metric is meaningful for a given row. Some legacy training billing events did not populate every metadata field. Missing values are returned as the proto default (empty string / 0) rather than as an error. securitySchemes: BearerAuth: type: http scheme: bearer description: >- Bearer authentication using your Fireworks API key. Format: Bearer bearerFormat: API_KEY ````