FlexAI Chat Completions API
Chat completions (streaming, tool use, vision).
Work with this as data
Every API here is available over the APIs.io API and to AI agents over MCP.
MCP server
One button, every client — Claude, Cursor, VS Code and the rest.
https://apis.io/mcp
Tools for apis
7 MCP tools reach this
find_apisBrowse and filter every API in the catalog.get_api_artifactsOne API's artifacts, grouped by type.get_openapiThe primary OpenAPI for this API.find_similar_apisAPIs that look like this one.apis_io_searchSTART HERE — APIs, providers and tags for one query, each with its total.resolveTurn a domain, URL or GitHub org into the provider it belongs to.find_cohortsEvery scored population of providers in the catalog.
Call it yourself
curl for this page
This API
curl "https://apis.io/api/v1/apis/flexai:flexai-chat-completions-api"
All apis
curl "https://apis.io/api/v1/apis?limit=25"Calls without X-API-Key need no key. Calls with it need a paid plan — get a key.
Get an API key
Free tier, no form to fill in. Signing in shares your email address with us — we store it to create your key and to recognise you if you sign in with another provider. See our Privacy Policy and Terms.
A second provider on the same verified email joins the account you already have.
OpenAPI Specification
openapi: 3.2.0
info:
title: Flexai Chat API
x-refined-note:
- x-provenance differs across the merged source definitions and was not carried
version: '1.0'
description: 'Operations tagged Chat across 2 of this provider''s published API definitions: flexai-openapi.yml, flexai-token-factory-openapi.yml. Each path carries the servers of the definition it was published in.'
servers:
- url: https://api.flex.ai
description: Production (US)
- url: https://eu.api.flex.ai
description: Production (EU)
- url: https://tokens.flex.ai/v1
description: FlexAI Token Factory (serverless inference)
security:
- bearerAuth: []
tags:
- name: Chat Completions
description: Chat completions (streaming, tool use, vision).
paths:
/v1/chat/completions:
post:
tags:
- Chat Completions
summary: Create a chat completion
description: 'Given a list of messages, returns a model-generated response. Supports
streaming (SSE), tool use, and (on some models) vision input.
**Streaming:** set `stream: true` and `stream_options: { include_usage: true }`
to receive token-by-token deltas. The final chunk (the one with `finish_reason`
set) carries the `usage` block; you need it for accurate cost tracking.'
operationId: createChatCompletion
requestBody:
required: true
content:
application/json:
schema:
type: object
required:
- model
- messages
properties:
model:
type: string
description: Model id. See `GET /v1/models` for available models.
example: Meta-Llama-3.1-8B-Instruct-FP8
messages:
type: array
items:
type: object
required:
- role
properties:
role:
type: string
enum:
- system
- user
- assistant
- tool
content:
oneOf:
- type: string
- type: array
items:
type: object
name:
type: string
tool_calls:
type: array
items:
type: object
tool_call_id:
type: string
stream:
type: boolean
default: false
stream_options:
type: object
properties:
include_usage:
type: boolean
description: Required for per-request spend tracking when streaming.
temperature:
type: number
minimum: 0
maximum: 2
default: 1
top_p:
type: number
minimum: 0
maximum: 1
max_tokens:
type: integer
minimum: 1
tools:
type: array
items:
type: object
required:
- type
- function
properties:
type:
type: string
enum:
- function
function:
type: object
required:
- name
properties:
name:
type: string
description:
type: string
parameters:
type: object
tool_choice:
oneOf:
- type: string
enum:
- none
- auto
- required
- type: object
response_format:
type: object
properties:
type:
type: string
enum:
- text
- json_object
seed:
type: integer
stop:
oneOf:
- type: string
- type: array
items:
type: string
user:
type: string
max_completion_tokens:
type: integer
minimum: 1
description: 'Newer name for max_tokens; either is honored as-is. Non-streaming requests that set neither default to 2048 output tokens on non-reasoning models (the response then carries `x-flexai-max-tokens-defaulted: 2048`); reasoning models get no default and are bounded by the non-streaming time budget instead.'
reasoning_effort:
type: string
enum:
- none
- minimal
- low
- medium
- high
- xhigh
- max
description: 'Accepted values differ by model. Models with a depth dial take the five gateway levels minimal, low, medium, high, max and return 400 for any other value, including none and xhigh: gpt-oss-120b/gpt-oss-20b accept low, medium, high only; DeepSeek-V4-Flash-0731 accepts all five (only minimal keeps it off; low, medium and high select the model''s high effort level, with low and medium reported as x-flexai-adjusted-params: reasoning_effort=high; max selects its maximum). A word that is no effort level at all is a 400 on every reasoning model; a model with no reasoning at all drops the parameter and reports x-flexai-dropped-params. Models with a reasoning on/off switch but no depth dial (gemma-4-31b-it, gemma-4-26B-A4B-it, Qwen3.5-9B, Qwen3.6-27B-FP8, Qwen3.6-35B-A3B-FP8, GLM-4.5-Air-FP8, Qwen3-8B-FP8, NVIDIA-Nemotron-3.5-Lightning-30B-A3B) take all seven values: low, medium, high, xhigh and max turn reasoning on at the model''s own depth and the response carries x-flexai-adjusted-params: reasoning_effort=on; none and minimal leave it off. An explicit thinking or chat_template_kwargs.enable_thinking switch wins over a contradicting effort, which is then reported in x-flexai-dropped-params. Reasoning is off by default on every model with an on/off switch; thinking {type: enabled} (or OpenRouter reasoning) turns it on everywhere.'
thinking:
type: object
description: On/off switch for reasoning on reasoning-capable models; takes precedence over reasoning_effort. disabled turns reasoning off even if reasoning_effort is sent; enabled alone turns it on at the model's default depth. Omitting both reasoning_effort and thinking leaves reasoning off on every switchable model. On a model that cannot stop reasoning (gpt-oss-120b, gpt-oss-20b and the always-thinking models) disabled is a 400 naming thinking, never a silent 200.
required:
- type
properties:
type:
type: string
enum:
- enabled
- disabled
budget_tokens:
type: integer
minimum: 1
responses:
'200':
description: 'Successful response. When `stream: true`, body is an SSE stream of `text/event-stream` chunks.'
headers:
x-flexai-max-tokens-defaulted:
schema:
type: integer
description: 'Present only when the request set neither `max_tokens` nor
`max_completion_tokens` and the gateway injected its default
output ceiling; the value is the ceiling applied. Only
non-streaming requests on non-reasoning models get the default,
so a reasoning model never carries this header.
'
x-flexai-max-tokens-clamped:
schema:
type: integer
description: 'Present only when the gateway lowered the request''s output
ceiling; the value is the ceiling applied. Reserved: no request
is clamped today. A STREAMED response the gateway ended early
(the model was repeating itself, or it reached the token
ceiling while repeating) is marked on its terminal chunks
instead, since its headers are already on the wire:
`finish_reason: "length"` plus `flexai_stopped: "degeneration"`
or `"ceiling"` (on `/v1/responses`, a `response.incomplete`
event with `incomplete_details.reason: "max_output_tokens"` and
the same field on the response object).
'
x-flexai-empty-at-length:
schema:
type: string
enum:
- 'true'
description: 'Present only on a non-streaming response whose first choice
ended with `finish_reason: "length"` and no visible content, no
`tool_calls` and no `function_call` — the model was cut off,
not incapable. Retry with a larger `max_tokens` or stream.
Never set on a streamed response: its headers are sent before
the terminal chunk exists.
'
x-flexai-dropped-params:
schema:
type: string
description: 'Present only when the gateway removed a request parameter the
model cannot use rather than rejecting the request; a
comma-separated list of the parameter names: `reasoning_effort`
on a model with no reasoning at all (or with an on/off switch
that the request also turned off), and `thinking` (or
`reasoning`, the OpenRouter dialect) on a model with no reasoning
switch, and `parallel_tool_calls` when a `tool_choice: "required"`
request sent `true` to a model the gateway bounds to one tool
call (`GLM-5.2`, `MiniMax-M2.7`,
`NVIDIA-Nemotron-3.5-Lightning-30B-A3B`), where it is forced to
`false`. A malformed value is a `400` on every model; only a
well-formed one the model cannot honour is dropped.
'
x-flexai-adjusted-params:
schema:
type: string
description: 'Present only when the gateway honoured a request parameter at
the nearest level the model supports rather than exactly; a
comma-separated list of `name=value`. Two cases today.
`reasoning_effort=on` (`reasoning=on` when sent as OpenRouter''s
`reasoning.effort` or on `/v1/responses`): the model has a
reasoning on/off switch but no depth setting, so any effort from
`low` to `max` turns reasoning on at the model''s own depth.
`none` and `minimal` turn it off and carry no header.
`reasoning_effort=high` (`reasoning=high` in the same spellings):
the model has two depths (`DeepSeek-V4-Flash-0731`), so a `low`
or `medium` was rounded up to its high level rather than
turning reasoning off, or a `minimal` was floored to it because
`thinking: {"type": "enabled"}` was also sent. `high`, `max` and
a bare `minimal` carry no header.
'
x-flexai-sampling-defaulted:
schema:
type: string
description: 'Present only when the gateway filled one or more sampling fields
from the model''s catalog `sampling_defaults` because the request
left them unset (absent or `null`); a comma-separated list of the
field names, e.g. `top_p`. A value the caller sent, including
`temperature: 0`, is never replaced. No header means nothing was
filled.
'
content:
application/json:
schema:
type: object
required:
- id
- object
- created
- model
- choices
properties:
id:
type: string
object:
type: string
enum:
- chat.completion
created:
type: integer
model:
type: string
choices:
type: array
items:
type: object
required:
- index
- message
- finish_reason
properties:
index:
type: integer
message:
type: object
properties:
role:
type: string
content:
type:
- string
- 'null'
reasoning_content:
type:
- string
- 'null'
description: Reasoning text when reasoning is enabled on a reasoning-capable model; absent or null otherwise.
tool_calls:
type: array
items:
type: object
finish_reason:
type: string
enum:
- stop
- length
- tool_calls
- content_filter
usage:
type: object
properties:
prompt_tokens:
type: integer
completion_tokens:
type: integer
total_tokens:
type: integer
completion_tokens_details:
type: object
description: 'Present when the model reasoned. reasoning_tokens are part of completion_tokens and billed as output; text_tokens is the visible-content remainder (live example: reasoning_tokens 78, text_tokens 2).'
properties:
reasoning_tokens:
type: integer
text_tokens:
type: integer
cache_read_input_tokens:
type: integer
cache_creation_input_tokens:
type: integer
cost:
type: number
description: 'FlexAI extension. Our charge for this request in USD, rounded
to 8 decimal places: what we bill you, not an upstream
provider''s cost. Equals
`cost_details.upstream_inference_prompt_cost` +
`cost_details.upstream_inference_completions_cost`. Present on
non-streaming responses and on the final streamed chunk''s
`usage` (with `stream_options.include_usage`). Absent when the
response cannot state the charge; the request is still billed,
and `/v1/flex/usage/request/{request_id}` returns the recorded
figure. Rates are published at `GET /v1/pricing`.
'
cost_details:
type: object
description: 'FlexAI extension. The parts of `cost`. Field names follow
OpenRouter''s usage-accounting fields for compatibility; the
values are still our charge to you, not a provider''s cost.
'
additionalProperties: false
properties:
upstream_inference_prompt_cost:
type: number
description: The input-token part of `cost`, in USD (cached-input rate for tokens served from the prompt cache).
upstream_inference_completions_cost:
type: number
description: The output-token part of `cost`, in USD, reasoning tokens included.
text/event-stream:
schema:
type: string
description: 'SSE stream. Each event is `data: <json>`. Terminates with `data: [DONE]`.'
'400':
$ref: '#/components/responses/BadRequest'
'401':
$ref: '#/components/responses/Unauthorized'
'402':
$ref: '#/components/responses/PaymentRequired'
'404':
$ref: '#/components/responses/NotFound'
'429':
$ref: '#/components/responses/RateLimited'
'500':
$ref: '#/components/responses/ServerError'
servers:
- url: https://api.flex.ai
description: Production (US)
- url: https://eu.api.flex.ai
description: Production (EU)
/chat/completions:
post:
operationId: postChatCompletions
summary: Create a chat completion
description: 'Generate a model response for a chat conversation. OpenAI-compatible. Supports streaming via server-sent events when stream=true. Note: non-streaming responses are capped at 2048 output tokens regardless of max_tokens (finish_reason "length"); n > 1 is rejected with a 400.'
tags:
- Chat Completions
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/ChatCompletionRequest'
responses:
'200':
description: A chat completion (or an SSE stream when stream=true)
content:
application/json:
schema:
$ref: '#/components/schemas/ChatCompletion'
'400':
$ref: '#/components/responses/BadRequest_2'
'401':
$ref: '#/components/responses/Unauthorized_2'
'429':
$ref: '#/components/responses/TooManyRequests'
x-operation-id-source: normalized
x-operation-id-original: createChatCompletion
servers:
- url: https://tokens.flex.ai/v1
description: FlexAI Token Factory (serverless inference)
components:
schemas:
Error:
type: object
required:
- error
properties:
error:
type: object
required:
- message
- type
properties:
message:
type: string
type:
type: string
example: invalid_request_error
code:
type: string
description: 'Stable machine-readable code for the failure. Present on
most error paths; OMITTED on FastAPI default-validation
bodies. Do not assume always present.
'
doc_url:
type: string
format: uri
description: 'FlexAI extension. Link to the relevant dashboard / docs
page for the failure mode (e.g. invalid key → /dashboard/keys).
Present on auth and quota errors.
'
ChatCompletionRequest:
type: object
required:
- model
- messages
properties:
model:
type: string
description: Model id from the /models catalog.
messages:
type: array
items:
$ref: '#/components/schemas/ChatMessage'
temperature:
type: number
top_p:
type: number
stop:
oneOf:
- type: string
- type: array
items:
type: string
seed:
type: integer
user:
type: string
max_tokens:
type: integer
description: Non-streaming responses are capped at 2048 output tokens regardless of this value.
stream:
type: boolean
stream_options:
type: object
properties:
include_usage:
type: boolean
tools:
type: array
items:
type: object
tool_choice:
oneOf:
- type: string
- type: object
response_format:
type: object
presence_penalty:
type: number
frequency_penalty:
type: number
ErrorResponse:
type: object
properties:
error:
type: object
properties:
message:
type: string
type:
type: string
param:
type:
- string
- 'null'
code:
type: string
doc_url:
type: string
ChatMessage:
type: object
required:
- role
- content
properties:
role:
type: string
enum:
- system
- user
- assistant
- tool
content:
type: string
name:
type: string
ChatCompletion:
type: object
properties:
id:
type: string
object:
type: string
example: chat.completion
created:
type: integer
model:
type: string
choices:
type: array
items:
type: object
properties:
index:
type: integer
message:
$ref: '#/components/schemas/ChatMessage'
finish_reason:
type: string
enum:
- stop
- length
- tool_calls
- content_filter
usage:
type: object
properties:
prompt_tokens:
type: integer
completion_tokens:
type: integer
total_tokens:
type: integer
responses:
Unauthorized:
description: Missing or invalid API key.
content:
application/json:
schema:
$ref: '#/components/schemas/Error'
NotFound:
description: Resource not found (e.g., unknown model or job id).
content:
application/json:
schema:
$ref: '#/components/schemas/Error'
RateLimited:
description: 'Per-key RPM exceeded. Response includes `Retry-After` and
`x-ratelimit-*` headers.
'
headers:
Retry-After:
schema:
type: integer
description: Seconds until the next window.
x-ratelimit-limit-requests:
schema:
type: integer
x-ratelimit-remaining-requests:
schema:
type: integer
x-ratelimit-reset-requests:
schema:
type: integer
content:
application/json:
schema:
$ref: '#/components/schemas/Error'
BadRequest:
description: Invalid request — malformed body, unknown model, or schema violation.
content:
application/json:
schema:
$ref: '#/components/schemas/Error'
PaymentRequired:
description: Key budget exhausted — add credit in the dashboard to continue.
content:
application/json:
schema:
$ref: '#/components/schemas/Error'
ServerError:
description: Upstream or gateway error. Safe to surface `error.message` to users.
content:
application/json:
schema:
$ref: '#/components/schemas/Error'
TooManyRequests:
description: Account-level rate limit or budget exceeded. Includes `Retry-After` and `x-ratelimit-*` response headers.
headers:
Retry-After:
schema:
type: integer
description: Seconds to wait before retrying.
content:
application/json:
schema:
$ref: '#/components/schemas/ErrorResponse'
Unauthorized_2:
description: Invalid or missing API key (error code `authentication_error`; includes a `doc_url` extension pointing at the auth docs).
content:
application/json:
schema:
$ref: '#/components/schemas/ErrorResponse'
BadRequest_2:
description: Request validation failure. The error object includes a `param` field indicating the offending path (OpenAI-compatible error envelope).
content:
application/json:
schema:
$ref: '#/components/schemas/ErrorResponse'
securitySchemes:
bearerAuth:
type: http
scheme: bearer
bearerFormat: sk-xxxx
description: 'Virtual API key. Create one from the
[FlexAI dashboard](https://platform.flex.ai/dashboard/keys). Pass as
`Authorization: Bearer sk-xxxx`.
'
x-refined-from:
- flexai-openapi.yml
- flexai-token-factory-openapi.yml