Skip to content
SayGMDocs
Dashboard

Create a message (Anthropic Messages API)

POST
/messages
curl --request POST \
--url https://api.saygm.com/v1/messages \
--header 'Content-Type: application/json' \
--header 'X-GM-Routing: default' \
--header 'anthropic-version: 2023-06-01' \
--header 'x-api-key: <x-api-key>' \
--data '{ "model": "claude-sonnet-4-6", "max_tokens": 1, "messages": [ { "role": "user", "content": [ { "type": "example", "text": "example", "source": {}, "id": "example", "name": "example", "input": {} } ] } ], "system": [ {} ], "stream": false, "temperature": 1, "top_p": 1, "top_k": 1, "stop_sequences": [ "example" ], "tools": [ {} ], "tool_choice": {}, "metadata": {} }'

Anthropic-compatible Messages API. The gateway requires Anthropic’s anthropic-version header and authenticates with the GM API key in the native x-api-key header. Set the Anthropic SDK base_url to https://api.saygm.com.

anthropic-version
required
string
>= 1 characters
Example
2023-06-01

Anthropic API version. The gateway rejects a missing or blank value before routing.

X-GM-Routing
string
default: default
Allowed values: default cheapest fastest

Per-request routing mode. default (or omitting the header) follows GM’s configured routing; cheapest ranks eligible routes by predicted final charge, preferring routes with at least 95% recent reliability; fastest ranks them by measured time to first token and output speed. Predictions are not a price cap or latency guarantee. A conversation stays on its current healthy route regardless of mode. Send the header at most once. Fusion, multi-model cascade, and some multimodal requests reject cheapest and fastest with 400.

X-GM-Providers
string
>= 1 characters

Comma-separated allowlist of provider identifiers that may serve the request. Use 1 to 16 distinct case-insensitive names, each 1 to 64 letters, digits, underscores, or hyphens, starting with a letter or digit. The restriction applies under every supported routing mode and on every retry. Invalid values return 400 with code invalid_provider_restriction; a model with no route on a listed provider returns 400 with code no_allowed_provider; listed providers that cannot serve now return 503. Fusion and multi-model cascade requests reject this header with 400.

Media typeapplication/json

The gateway requires model and messages. Additional provider fields are forwarded unchanged.

object
model
required

Canonical buyer model id from /v1/models.

string
max_tokens

Optional maximum output token count. When omitted, the gateway supplies an output bound using published model limits or its configured default.

integer
>= 1
messages
required
Array<object>
>= 1 items
object
role
required
string
Allowed values: user assistant system
content
required
One of:
string
key
additional properties
any
system
One of:
string
stream
boolean
temperature
number
<= 1
top_p
number
<= 1
top_k
integer
stop_sequences
Array<string>
tools
Array<object>
object
tool_choice
object
metadata
object
key
additional properties
any

An Anthropic Messages object, or a text/event-stream when stream is true.

object
id
required
string
type
required
string
Allowed value: message
role
required
string
Allowed value: assistant
model
required
string
content
required
Array<object>
object
type
string
text
string
source
object
id
string
name
string
input
object
key
additional properties
any
stop_reason
string | null
stop_sequence
string | null
usage
required
object
input_tokens
integer
output_tokens
integer
key
additional properties
any
key
additional properties
any
Example
{
"type": "message",
"role": "assistant"
}

Invalid request or an upstream provider 400 response.

Media typeapplication/json
object
type
required
string
Allowed value: error
error
required
object
type
required
string
message
required
string
key
additional properties
any
request_id

Gateway request identifier.

string
key
additional properties
any
Example
{
"type": "error"
}

The API key is missing or invalid.

Media typeapplication/json
object
type
required
string
Allowed value: error
error
required
object
type
required
string
message
required
string
key
additional properties
any
request_id

Gateway request identifier.

string
key
additional properties
any
Example
{
"type": "error"
}

Insufficient prepaid balance. The error message links to the funding page; when the available balance is nonzero it also reports estimated maximum cost and request-sizing guidance.

Media typeapplication/json
object
type
required
string
Allowed value: error
error
required
object
type
required
string
message
required
string
key
additional properties
any
request_id

Gateway request identifier.

string
key
additional properties
any
Examples

No available balance

{
"type": "error",
"error": {
"type": "permission_error",
"message": "Your gm balance is empty. Add credits: https://saygm.com/topup"
}
}

Guardrail block (guardrail_blocked).

Media typeapplication/json
Any of:

OpenAI-shaped gateway error. Image errors after dispatch may also include billing fields at the top level.

object
error
required
object
message
required
string
type
required
string
code
required

Machine-readable error code. Account request limits use rate_limit_exceeded.

string
param
required
string | null
request_id

Gateway request identifier.

string
charged
boolean
billing_status
string
Allowed values: settled unbilled unknown
potentially_charged
boolean
cost_ndollars

Settled charge in nano-dollars; present when billing_status is settled.

integer
Example
{
"billing_status": "settled"
}

Request body too large (request_too_large).

Media typeapplication/json
Any of:

OpenAI-shaped gateway error. Image errors after dispatch may also include billing fields at the top level.

object
error
required
object
message
required
string
type
required
string
code
required

Machine-readable error code. Account request limits use rate_limit_exceeded.

string
param
required
string | null
request_id

Gateway request identifier.

string
charged
boolean
billing_status
string
Allowed values: settled unbilled unknown
potentially_charged
boolean
cost_ndollars

Settled charge in nano-dollars; present when billing_status is settled.

integer
Example
{
"billing_status": "settled"
}

Upstream 422 response.

Media typeapplication/json
Any of:

OpenAI-shaped gateway error. Image errors after dispatch may also include billing fields at the top level.

object
error
required
object
message
required
string
type
required
string
code
required

Machine-readable error code. Account request limits use rate_limit_exceeded.

string
param
required
string | null
request_id

Gateway request identifier.

string
charged
boolean
billing_status
string
Allowed values: settled unbilled unknown
potentially_charged
boolean
cost_ndollars

Settled charge in nano-dollars; present when billing_status is settled.

integer
Example
{
"billing_status": "settled"
}

Account, model, or upstream request limit. OpenAI-shaped account-limit errors use rate_limit_exceeded.

Media typeapplication/json
Any of:

OpenAI-shaped gateway error. Image errors after dispatch may also include billing fields at the top level.

object
error
required
object
message
required
string
type
required
string
code
required

Machine-readable error code. Account request limits use rate_limit_exceeded.

string
param
required
string | null
request_id

Gateway request identifier.

string
charged
boolean
billing_status
string
Allowed values: settled unbilled unknown
potentially_charged
boolean
cost_ndollars

Settled charge in nano-dollars; present when billing_status is settled.

integer
Example
{
"billing_status": "settled"
}
Retry-After
integer
>= 1

Seconds before retrying, when a delay is available.

Gateway internal error (internal_error).

Media typeapplication/json
Any of:

OpenAI-shaped gateway error. Image errors after dispatch may also include billing fields at the top level.

object
error
required
object
message
required
string
type
required
string
code
required

Machine-readable error code. Account request limits use rate_limit_exceeded.

string
param
required
string | null
request_id

Gateway request identifier.

string
charged
boolean
billing_status
string
Allowed values: settled unbilled unknown
potentially_charged
boolean
cost_ndollars

Settled charge in nano-dollars; present when billing_status is settled.

integer
Example
{
"billing_status": "settled"
}

Upstream connection failure (upstream_connect_failed).

Media typeapplication/json
Any of:

OpenAI-shaped gateway error. Image errors after dispatch may also include billing fields at the top level.

object
error
required
object
message
required
string
type
required
string
code
required

Machine-readable error code. Account request limits use rate_limit_exceeded.

string
param
required
string | null
request_id

Gateway request identifier.

string
charged
boolean
billing_status
string
Allowed values: settled unbilled unknown
potentially_charged
boolean
cost_ndollars

Settled charge in nano-dollars; present when billing_status is settled.

integer
Example
{
"billing_status": "settled"
}

The credit service, guardrail policy, or model serving is unavailable.

Media typeapplication/json
Any of:

OpenAI-shaped gateway error. Image errors after dispatch may also include billing fields at the top level.

object
error
required
object
message
required
string
type
required
string
code
required

Machine-readable error code. Account request limits use rate_limit_exceeded.

string
param
required
string | null
request_id

Gateway request identifier.

string
charged
boolean
billing_status
string
Allowed values: settled unbilled unknown
potentially_charged
boolean
cost_ndollars

Settled charge in nano-dollars; present when billing_status is settled.

integer
Example
{
"billing_status": "settled"
}
Retry-After
integer
>= 1

Seconds before retrying, when a delay is available.

Upstream timeout (upstream_timeout) or idle stream timeout (upstream_stream_idle_timeout).

Media typeapplication/json
Any of:

OpenAI-shaped gateway error. Image errors after dispatch may also include billing fields at the top level.

object
error
required
object
message
required
string
type
required
string
code
required

Machine-readable error code. Account request limits use rate_limit_exceeded.

string
param
required
string | null
request_id

Gateway request identifier.

string
charged
boolean
billing_status
string
Allowed values: settled unbilled unknown
potentially_charged
boolean
cost_ndollars

Settled charge in nano-dollars; present when billing_status is settled.

integer
Example
{
"billing_status": "settled"
}