Fireworks AI Streaming Inference API

asyncapi: '2.6.0'
id: 'urn:com:fireworks:ai:inference:streaming'
info:
  title: Fireworks AI Streaming Inference API
  version: '1.0.0'
  description: |
    AsyncAPI description of the Fireworks AI streaming inference surface. Fireworks
    streams generation deltas over HTTP using Server-Sent Events (SSE) on a single
    `text/event-stream` response when `stream: true` is set on the request body.

    Only endpoints that Fireworks AI documents as supporting SSE streaming are
    described here:

      - POST /chat/completions  (OpenAI-compatible chat completions stream)
      - POST /completions       (OpenAI-compatible legacy text completions stream)
      - POST /responses         (OpenAI-compatible Responses API stream)
      - POST /messages          (Anthropic-compatible Messages stream)

    Fireworks AI does not document a separate streaming endpoint for audio
    transcription, translation, or TTS. Audio is delivered to the platform as an
    `audio_url` content part inside a chat completion request and is streamed back
    using the same `/chat/completions` SSE stream described below.

    Sources:
      - https://docs.fireworks.ai/api-reference/post-chatcompletions
      - https://docs.fireworks.ai/api-reference/post-completions
      - https://docs.fireworks.ai/api-reference/post-responses
      - https://docs.fireworks.ai/api-reference/anthropic-messages
      - https://docs.fireworks.ai/guides/querying-text-models
      - https://docs.fireworks.ai/guides/function-calling
      - https://docs.fireworks.ai/guides/video-audio-inputs
  contact:
    name: Fireworks AI
    url: https://docs.fireworks.ai/
  license:
    name: Proprietary
    url: https://fireworks.ai/terms-of-service
  tags:
    - name: Streaming
    - name: SSE
    - name: LLM
    - name: Inference

defaultContentType: text/event-stream

servers:
  production:
    url: https://api.fireworks.ai/inference/v1
    protocol: https
    description: |
      Fireworks AI inference base URL. All streaming endpoints are reached by
      sending an HTTP POST with `stream: true` (or `"stream": true`) in the JSON
      body; the server responds with `Content-Type: text/event-stream` and emits
      a sequence of `data:` lines terminated by `data: [DONE]`.
    security:
      - bearerAuth: []
    bindings:
      http:
        bindingVersion: '0.3.0'

channels:

  /chat/completions:
    description: |
      OpenAI-compatible chat completions. When the request body sets
      `stream: true`, the response is `text/event-stream`. Each event is emitted
      as a `data:` line whose payload is a JSON `ChatCompletionStreamResponse`
      chunk. The stream terminates with a literal `data: [DONE]` event.
    bindings:
      http:
        type: request
        method: POST
        bindingVersion: '0.3.0'
    subscribe:
      operationId: streamChatCompletion
      summary: Receive chat completion deltas as Server-Sent Events.
      description: |
        Token-by-token deltas of a chat completion. The final non-terminator
        chunk carries `finish_reason`, optional `usage`, and (when requested)
        `perf_metrics`. The stream is closed by a `data: [DONE]` line.
      message:
        oneOf:
          - $ref: '#/components/messages/ChatCompletionChunk'
          - $ref: '#/components/messages/StreamDone'
    publish:
      operationId: createChatCompletionStream
      summary: Open a chat completion stream.
      description: |
        POST a `ChatCompletionRequest` with `stream: true` to open the SSE
        stream. The same body schema applies whether or not streaming is used.
      message:
        $ref: '#/components/messages/ChatCompletionRequest'

  /completions:
    description: |
      OpenAI-compatible legacy text completions. When the request body sets
      `stream: true`, the response is `text/event-stream`. Each event is a
      `data:` line whose payload is a JSON `CompletionStreamResponse` chunk.
      The stream terminates with `data: [DONE]`.
    bindings:
      http:
        type: request
        method: POST
        bindingVersion: '0.3.0'
    subscribe:
      operationId: streamTextCompletion
      summary: Receive text completion deltas as Server-Sent Events.
      description: |
        Token deltas for the legacy completions endpoint. The final non-terminator
        chunk carries `finish_reason`, optional `usage`, and (when requested)
        `perf_metrics`. The stream is closed by a `data: [DONE]` line.
      message:
        oneOf:
          - $ref: '#/components/messages/CompletionChunk'
          - $ref: '#/components/messages/StreamDone'
    publish:
      operationId: createTextCompletionStream
      summary: Open a text completion stream.
      message:
        $ref: '#/components/messages/CompletionRequest'

  /responses:
    description: |
      OpenAI-compatible Responses API. When the request body sets `stream: true`,
      the response is `text/event-stream`. Per Fireworks docs each chunk is an
      SSE event delivering the incremental Response state. Fireworks does not
      enumerate the full set of event names in public documentation; the
      generic event payload is described here.
    bindings:
      http:
        type: request
        method: POST
        bindingVersion: '0.3.0'
    subscribe:
      operationId: streamResponse
      summary: Receive Response streaming events as Server-Sent Events.
      description: |
        Server-Sent Events emitted while a Response is being generated. Each
        event payload is a partial or final Response object. The stream closes
        once the Response reaches a terminal status (e.g. `completed`,
        `failed`, `incomplete`, `cancelled`).
      message:
        $ref: '#/components/messages/ResponseStreamEvent'
    publish:
      operationId: createResponseStream
      summary: Open a Response stream.
      message:
        $ref: '#/components/messages/ResponseRequest'

  /messages:
    description: |
      Anthropic-compatible Messages endpoint. When `stream: true`, the response
      is `text/event-stream`. Unlike the OpenAI-compatible endpoints, each SSE
      event includes both an `event:` line naming the event type and a `data:`
      line carrying the JSON payload. Event types are enumerated below.
    bindings:
      http:
        type: request
        method: POST
        bindingVersion: '0.3.0'
    subscribe:
      operationId: streamAnthropicMessage
      summary: Receive Anthropic-compatible Message streaming events.
      description: |
        A documented sequence of typed SSE events: `message_start` opens the
        stream with an initial Message envelope, one or more
        `content_block_start` / `content_block_delta` / `content_block_stop`
        groups deliver content blocks, `message_delta` carries top-level
        updates such as `stop_reason`, and `message_stop` closes the stream.
      message:
        oneOf:
          - $ref: '#/components/messages/AnthropicMessageStart'
          - $ref: '#/components/messages/AnthropicContentBlockStart'
          - $ref: '#/components/messages/AnthropicContentBlockDelta'
          - $ref: '#/components/messages/AnthropicContentBlockStop'
          - $ref: '#/components/messages/AnthropicMessageDelta'
          - $ref: '#/components/messages/AnthropicMessageStop'
    publish:
      operationId: createAnthropicMessageStream
      summary: Open an Anthropic-compatible Messages stream.
      message:
        $ref: '#/components/messages/AnthropicMessageRequest'

components:

  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer
      bearerFormat: API Key
      description: |
        Fireworks AI API key passed as `Authorization: Bearer <FIREWORKS_API_KEY>`.

  messages:

    # ---------- Chat Completions ----------

    ChatCompletionRequest:
      name: ChatCompletionRequest
      title: Chat Completion Request
      contentType: application/json
      summary: Body of POST /chat/completions with stream=true.
      payload:
        $ref: '#/components/schemas/ChatCompletionRequest'
      bindings:
        http:
          headers:
            type: object
            properties:
              Authorization:
                type: string
                description: 'Bearer <FIREWORKS_API_KEY>'
              Content-Type:
                type: string
                const: application/json
              Accept:
                type: string
                const: text/event-stream
          bindingVersion: '0.3.0'

    ChatCompletionChunk:
      name: ChatCompletionChunk
      title: Chat Completion Stream Chunk
      contentType: application/json
      summary: A single `data:` SSE event carrying an incremental delta.
      description: |
        SSE event of the form `data: {ChatCompletionStreamResponse}\n\n`. The
        terminal stream marker `data: [DONE]` is described by `StreamDone`.
      payload:
        $ref: '#/components/schemas/ChatCompletionStreamResponse'
      examples:
        - name: typical-token-chunk
          summary: A typical mid-stream token chunk
          payload:
            id: cmpl-xyz
            object: chat.completion.chunk
            created: 1748501234
            model: accounts/fireworks/models/kimi-k2-instruct-0905
            choices:
              - index: 0
                delta:
                  content: 'Hello'
                finish_reason: null

    # ---------- Completions ----------

    CompletionRequest:
      name: CompletionRequest
      title: Text Completion Request
      contentType: application/json
      summary: Body of POST /completions with stream=true.
      payload:
        $ref: '#/components/schemas/CompletionRequest'
      bindings:
        http:
          headers:
            type: object
            properties:
              Authorization:
                type: string
                description: 'Bearer <FIREWORKS_API_KEY>'
              Content-Type:
                type: string
                const: application/json
              Accept:
                type: string
                const: text/event-stream
          bindingVersion: '0.3.0'

    CompletionChunk:
      name: CompletionChunk
      title: Text Completion Stream Chunk
      contentType: application/json
      summary: A single `data:` SSE event carrying a token delta.
      payload:
        $ref: '#/components/schemas/CompletionStreamResponse'

    # ---------- Responses ----------

    ResponseRequest:
      name: ResponseRequest
      title: Responses API Request
      contentType: application/json
      summary: Body of POST /responses with stream=true.
      payload:
        $ref: '#/components/schemas/ResponseRequest'
      bindings:
        http:
          headers:
            type: object
            properties:
              Authorization:
                type: string
                description: 'Bearer <FIREWORKS_API_KEY>'
              Content-Type:
                type: string
                const: application/json
              Accept:
                type: string
                const: text/event-stream
          bindingVersion: '0.3.0'

    ResponseStreamEvent:
      name: ResponseStreamEvent
      title: Response Stream Event
      contentType: application/json
      summary: An SSE event delivering an incremental Response object.
      description: |
        Per Fireworks docs, when `stream: true`, the Responses API "returns
        responses via Server-Sent Events (SSE), delivering tokens incrementally
        as they are generated." Fireworks public docs reference event names
        such as `response.created`, `response.in_progress`,
        `response.output_text.delta`, and `response.completed`, and direct
        readers to the streaming cookbook for the full event catalogue.
      payload:
        $ref: '#/components/schemas/ResponseStreamPayload'

    # ---------- Anthropic Messages ----------

    AnthropicMessageRequest:
      name: AnthropicMessageRequest
      title: Anthropic Messages Request
      contentType: application/json
      summary: Body of POST /messages with stream=true.
      payload:
        $ref: '#/components/schemas/AnthropicMessageRequest'
      bindings:
        http:
          headers:
            type: object
            properties:
              Authorization:
                type: string
                description: 'Bearer <FIREWORKS_API_KEY>'
              Content-Type:
                type: string
                const: application/json
              Accept:
                type: string
                const: text/event-stream
          bindingVersion: '0.3.0'

    AnthropicMessageStart:
      name: AnthropicMessageStart
      title: message_start
      contentType: application/json
      summary: Opens an Anthropic-compatible message stream.
      payload:
        $ref: '#/components/schemas/AnthropicMessageStartEvent'

    AnthropicContentBlockStart:
      name: AnthropicContentBlockStart
      title: content_block_start
      contentType: application/json
      summary: Announces the start of a content block in the Message.
      payload:
        $ref: '#/components/schemas/AnthropicContentBlockStartEvent'

    AnthropicContentBlockDelta:
      name: AnthropicContentBlockDelta
      title: content_block_delta
      contentType: application/json
      summary: Incremental content for the active content block.
      payload:
        $ref: '#/components/schemas/AnthropicContentBlockDeltaEvent'

    AnthropicContentBlockStop:
      name: AnthropicContentBlockStop
      title: content_block_stop
      contentType: application/json
      summary: Marks the end of a content block.
      payload:
        $ref: '#/components/schemas/AnthropicContentBlockStopEvent'

    AnthropicMessageDelta:
      name: AnthropicMessageDelta
      title: message_delta
      contentType: application/json
      summary: Top-level Message updates (e.g., stop_reason, usage).
      payload:
        $ref: '#/components/schemas/AnthropicMessageDeltaEvent'

    AnthropicMessageStop:
      name: AnthropicMessageStop
      title: message_stop
      contentType: application/json
      summary: Terminates the Anthropic Messages SSE stream.
      payload:
        $ref: '#/components/schemas/AnthropicMessageStopEvent'

    # ---------- Stream terminator ----------

    StreamDone:
      name: StreamDone
      title: '[DONE] terminator'
      contentType: text/plain
      summary: 'Final SSE line `data: [DONE]` closing an OpenAI-compatible stream.'
      description: |
        Per Fireworks docs the OpenAI-compatible chat and completions streams
        terminate with the literal SSE line `data: [DONE]`.
      payload:
        type: string
        const: '[DONE]'

  schemas:

    # ---------- Chat Completions schemas ----------

    ChatCompletionRequest:
      type: object
      required: [model, messages]
      properties:
        model:
          type: string
          description: 'Model identifier, e.g. accounts/fireworks/models/kimi-k2-instruct-0905.'
        messages:
          type: array
          items:
            $ref: '#/components/schemas/ChatMessage'
        stream:
          type: boolean
          default: false
          description: When true, response is delivered as Server-Sent Events.
        tools:
          type: array
          items:
            $ref: '#/components/schemas/ChatCompletionTool'
        tool_choice:
          oneOf:
            - type: string
              enum: [auto, none, required, any]
            - type: object
        parallel_tool_calls:
          type: boolean
        functions:
          type: array
          description: Deprecated; legacy function definitions.
          items:
            type: object
        function_call:
          description: Deprecated; use tool_choice.
        temperature:
          type: number
          minimum: 0
          maximum: 2
        top_p:
          type: number
          minimum: 0
          maximum: 1
        top_k:
          type: integer
          minimum: 0
          maximum: 100
        min_p:
          type: number
          minimum: 0
          maximum: 1
        typical_p:
          type: number
          minimum: 0
          maximum: 1
        frequency_penalty:
          type: number
          minimum: -2
          maximum: 2
        presence_penalty:
          type: number
          minimum: -2
          maximum: 2
        repetition_penalty:
          type: number
          minimum: 0
          maximum: 2
        max_tokens:
          type: integer
        max_completion_tokens:
          type: integer
        stop:
          oneOf:
            - type: string
            - type: array
              items:
                type: string
              maxItems: 4
        response_format:
          type: object
          properties:
            type:
              type: string
              enum: [json_object, json_schema, grammar, text]
        reasoning_effort:
          oneOf:
            - type: string
              enum: [low, medium, high, xhigh, max, none]
            - type: integer
              minimum: 1024
        reasoning_history:
          type: string
          enum: [disabled, interleaved, preserved]
        thinking:
          type: object
        prompt_cache_key:
          type: string
        prompt_cache_isolation_key:
          type: string
        prompt_truncate_len:
          type: integer
        safe_tokenization:
          type: boolean
        logprobs:
          oneOf:
            - type: boolean
            - type: integer
              minimum: 0
              maximum: 5
        top_logprobs:
          type: integer
          minimum: 0
          maximum: 5
        echo:
          type: boolean
        echo_last:
          type: integer
        return_token_ids:
          type: boolean
        raw_output:
          type: boolean
        perf_metrics_in_response:
          type: boolean
        speculation:
          oneOf:
            - type: string
            - type: array
              items:
                type: integer
        prediction:
          oneOf:
            - type: object
            - type: string
        seed:
          type: integer
        user:
          type: string
        metadata:
          type: object
        service_tier:
          type: string
          enum: [auto, default, flex, priority]
        ignore_eos:
          type: boolean
        context_length_exceeded_behavior:
          type: string
          enum: [truncate, error]
        logit_bias:
          type: object
        n:
          type: integer
          minimum: 1
          maximum: 128
        mirostat_target:
          type: number
        mirostat_lr:
          type: number

    ChatMessage:
      type: object
      required: [role]
      properties:
        role:
          type: string
          enum: [system, user, assistant, tool]
        content:
          oneOf:
            - type: string
            - type: array
              items:
                $ref: '#/components/schemas/ChatMessageContent'
        reasoning_content:
          type: string
        tool_calls:
          type: array
          items:
            $ref: '#/components/schemas/ToolCall'
        tool_call_id:
          type: string

    ChatMessageContent:
      type: object
      description: |
        Multimodal content part. Vision uses `image_url`; video and audio use
        `video_url` and `audio_url` (audio as a base64 data URL, e.g.
        `data:audio/ogg;base64,...`).
      properties:
        type:
          type: string
          enum: [text, image_url, video_url, audio_url]
        text:
          type: string
        image_url:
          type: object
          properties:
            url:
              type: string
        video_url:
          type: object
          properties:
            url:
              type: string
        audio_url:
          type: object
          properties:
            url:
              type: string
              description: 'Base64 data URL, e.g. data:audio/ogg;base64,<DATA>.'

    ChatCompletionTool:
      type: object
      required: [type, function]
      properties:
        type:
          type: string
          const: function
        function:
          type: object
          required: [name]
          properties:
            name:
              type: string
            description:
              type: string
            parameters:
              type: object
              description: JSON Schema for function arguments.

    ToolCall:
      type: object
      properties:
        id:
          type: string
        type:
          type: string
          const: function
        function:
          type: object
          properties:
            name:
              type: string
            arguments:
              type: string
              description: JSON-encoded arguments string.

    ChatCompletionStreamResponse:
      type: object
      description: |
        Payload of one `data:` SSE event during a chat completions stream.
      required: [id, object, created, model, choices]
      properties:
        id:
          type: string
        object:
          type: string
          const: chat.completion.chunk
        created:
          type: integer
          description: Unix timestamp.
        model:
          type: string
        choices:
          type: array
          items:
            $ref: '#/components/schemas/ChatCompletionStreamChoice'
        usage:
          $ref: '#/components/schemas/UsageInfo'
        perf_metrics:
          $ref: '#/components/schemas/PerfMetrics'
        prompt_token_ids:
          type: array
          items:
            type: integer

    ChatCompletionStreamChoice:
      type: object
      required: [index, delta]
      properties:
        index:
          type: integer
        delta:
          $ref: '#/components/schemas/ChatCompletionDelta'
        finish_reason:
          oneOf:
            - type: 'null'
            - type: string
              enum: [stop, length, function_call, tool_calls]
        logprobs:
          oneOf:
            - type: 'null'
            - type: object
        raw_output:
          type: object
        prompt_token_ids:
          type: array
          items:
            type: integer
        token_ids:
          type: array
          items:
            type: integer

    ChatCompletionDelta:
      type: object
      properties:
        role:
          type: string
        content:
          type: string
        reasoning_content:
          type: string
        tool_calls:
          type: array
          items:
            $ref: '#/components/schemas/ToolCallDelta'

    ToolCallDelta:
      type: object
      properties:
        index:
          type: integer
        id:
          type: string
        type:
          type: string
          const: function
        function:
          type: object
          properties:
            name:
              type: string
            arguments:
              type: string
              description: |
                Incremental JSON string fragment; clients accumulate fragments
                across chunks until `finish_reason == "tool_calls"`.

    # ---------- Completions schemas ----------

    CompletionRequest:
      type: object
      required: [model, prompt]
      properties:
        model:
          type: string
        prompt:
          oneOf:
            - type: string
            - type: array
              items:
                type: string
            - type: array
              items:
                type: integer
            - type: array
              items:
                type: array
                items:
                  type: integer
        stream:
          type: boolean
          default: false
        max_tokens:
          type: integer
        max_completion_tokens:
          type: integer
        temperature:
          type: number
          minimum: 0
          maximum: 2
        top_p:
          type: number
          minimum: 0
          maximum: 1
        top_k:
          type: integer
          minimum: 0
          maximum: 100
        top_logprobs:
          type: integer
          minimum: 0
          maximum: 5
        stop:
          oneOf:
            - type: string
            - type: array
              items:
                type: string
        logprobs:
          oneOf:
            - type: boolean
            - type: integer
              minimum: 0
              maximum: 5
        echo:
          type: boolean
        n:
          type: integer
          minimum: 1
          maximum: 128
        response_format:
          type: object
        reasoning_effort:
          oneOf:
            - type: string
              enum: [low, medium, high, xhigh, max, none]
            - type: integer
        thinking:
          type: object
        min_p:
          type: number
        typical_p:
          type: number
        frequency_penalty:
          type: number
        presence_penalty:
          type: number
        repetition_penalty:
          type: number
        mirostat_target:
          type: number
        mirostat_lr:
          type: number

    CompletionStreamResponse:
      type: object
      description: Payload of one `data:` SSE event during a text completions stream.
      required: [id, object, created, model, choices]
      properties:
        id:
          type: string
        object:
          type: string
          const: text_completion
        created:
          type: integer
        model:
          type: string
        choices:
          type: array
          items:
            $ref: '#/components/schemas/CompletionStreamChoice'
        usage:
          oneOf:
            - type: 'null'
            - $ref: '#/components/schemas/UsageInfo'
        perf_metrics:
          oneOf:
            - type: 'null'
            - $ref: '#/components/schemas/PerfMetrics'

    CompletionStreamChoice:
      type: object
      required: [index, text]
      properties:
        index:
          type: integer
        text:
          type: string
        finish_reason:
          oneOf:
            - type: 'null'
            - type: string
              enum: [stop, length, error]
        token_ids:
          type: array
          items:
            type: integer

    # ---------- Responses schemas ----------

    ResponseRequest:
      type: object
      required: [model, input]
      properties:
        model:
          type: string
        input:
          oneOf:
            - type: string
            - type: array
              items:
                type: object
        previous_response_id:
          type: string
        instructions:
          type: string
        max_output_tokens:
          type: integer
          minimum: 1
        max_tool_calls:
          type: integer
          minimum: 1
        metadata:
          type: object
        parallel_tool_calls:
          type: boolean
          default: true
        reasoning:
          type: object
        store:
          type: boolean
          default: true
        stream:
          type: boolean
          default: false
        temperature:
          type: number
          minimum: 0
          maximum: 2
        tool_choice:
          oneOf:
            - type: string
              enum: [none, auto, required]
            - type: object
        tools:
          type: array
          items:
            type: object
            description: |
              Supports `function`, `mcp`, `sse`, and `python` tool types.
        top_p:
          type: number
          minimum: 0
          maximum: 1
        truncation:
          type: string
          enum: [auto, disabled]
          default: disabled
        user:
          type: string
        text:
          type: object

    ResponseStreamPayload:
      type: object
      description: |
        Generic Response stream event payload. Each SSE event delivers a
        partial or final Response object whose `status` advances through
        values such as `in_progress` and a terminal state (`completed`,
        `failed`, `incomplete`, or `cancelled`).
      properties:
        type:
          type: string
          description: |
            Event type name. Fireworks docs reference event names of the form
            `response.created`, `response.in_progress`,
            `response.output_text.delta`, and `response.completed`. The
            complete event taxonomy is provided by the Fireworks streaming
            cookbook rather than the public API reference.
        response:
          type: object
          description: Full or partial Response envelope at this point in the stream.
        delta:
          description: Incremental payload for delta-style events.

    # ---------- Anthropic Messages schemas ----------

    AnthropicMessageRequest:
      type: object
      required: [model, messages, max_tokens]
      properties:
        model:
          type: string
        messages:
          type: array
          items:
            type: object
        max_tokens:
          type: integer
          minimum: 1
        system:
          oneOf:
            - type: string
            - type: array
              items:
                type: object
        temperature:
          type: number
          minimum: 0
          maximum: 1
        top_p:
          type: number
          minimum: 0
          maximum: 1
        top_k:
          type: integer
          minimum: 0
        stop_sequences:
          type: array
          items:
            type: string
        stream:
          type: boolean
        metadata:
          type: object
        output_config:
          type: object
        tool_choice:
          oneOf:
            - type: string
              enum: [auto, any, none]
            - type: object
        tools:
          type: array
          items:
            type: object
        thinking:
          type: object
        raw_output:
          type: boolean

    AnthropicMessageStartEvent:
      type: object
      required: [type, message]
      properties:
        type:
          type: string
          const: message_start
        message:
          type: object
          description: Initial Message envelope with id, role, model, and empty content array.

    AnthropicContentBlockStartEvent:
      type: object
      required: [type, index, content_block]
      properties:
        type:
          type: string
          const: content_block_start
        index:
          type: integer
        content_block:
          type: object

    AnthropicContentBlockDeltaEvent:
      type: object
      required: [type, index, delta]
      properties:
        type:
          type: string
          const: content_block_delta
        index:
          type: integer
        delta:
          type: object

    AnthropicContentBlockStopEvent:
      type: object
      required: [type, index]
      properties:
        type:
          type: string
          const: content_block_stop
        index:
          type: integer

    AnthropicMessageDeltaEvent:
      type: object
      required: [type, delta]
      properties:
        type:
          type: string
          const: message_delta
        delta:
          type: object
          description: |
            Top-level Message updates. `stop_reason` may be one of `end_turn`,
            `max_tokens`, `stop_sequence`, `tool_use`, `pause_turn`, or
            `refusal`.
        usage:
          type: object

    AnthropicMessageStopEvent:
      type: object
      required: [type]
      properties:
        type:
          type: string
          const: message_stop

    # ---------- Shared schemas ----------

    UsageInfo:
      type: object
      properties:
        prompt_tokens:
          type: integer
        completion_tokens:
          oneOf:
            - type: 'null'
            - type: integer
        total_tokens:
          type: integer
        prompt_tokens_details:
          type: object
          properties:
            cached_tokens:
              oneOf:
                - type: 'null'
                - type: integer

    PerfMetrics:
      type: object
      description: |
        Performance metrics returned in the final stream chunk when
        `perf_metrics_in_response=true`. For dedicated deployments includes
        deployment, queue, and speculative-decoding metrics.
      properties:
        prompt-tokens:
          type: integer
        cached-prompt-tokens:
          type: integer
        server-time-to-first-token:
          type: number
        server-processing-time:
          type: number
        speculation-prompt-tokens:
          type: integer
        speculation-prompt-matched-tokens:
         

# --- truncated at 32 KB (32 KB total) ---
# Full source: https://raw.githubusercontent.com/api-evangelist/fireworks-ai/refs/heads/main/asyncapi/fireworks-ai-asyncapi.yml