gpt-4o-mini

openai/gpt-4o-mini — a llm model by OpenAI. Call it through the gateway’s standard OpenAI-compatible door using the catalog slug; the full parameter schema is reproduced in full below.

Catalog slug	openai/gpt-4o-mini
Kind	llm
Vendor	OpenAI
Native model id	gpt-4o-mini

How to call

Through the Sociaro SDK / OpenAI-compatible door

from sociaro_ai import Sociaro

client = Sociaro(api_key="gw_live_...")   # base_url defaults to https://api.sociaro.com

resp = client.chat.completions.create(
    model="openai/gpt-4o-mini",          # always the catalog slug
    messages=[{"role": "user", "content": "Hello"}],
    # any parameter from the schema below is passed through unchanged
)
print(resp.choices[0].message.content)

TypeScript: client.chat.completions.create({ model: "openai/gpt-4o-mini", messages: [...] }) — see the TypeScript SDK.

Through OpenAI’s native API (advanced)

Optional: send OpenAI’s native request to the /openai prefix, using the native model id gpt-4o-mini. The gateway injects the upstream OpenAI credential and forwards the body unchanged (catalog, pricing and entitlements still apply).

curl https://api.sociaro.com/openai/<native-path> \
  -H "Authorization: Bearer $SOCIARO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "model": "gpt-4o-mini", ... }'

The exact path, request body and response shape are documented in full in the schema below.

Parameters — full schema

The complete, one-to-one API schema for this model, transcribed from the official source. Nothing omitted. Examples in the schema may use a representative model of the family — for this model use the slug openai/gpt-4o-mini (or the native id gpt-4o-mini for the native API) as shown in How to call above.

Coverage map (models in this family)

author slug	reasoning model?	text	image input	audio input	audio output	web_search_options	notes
openai/gpt-4o	no	yes	yes (image_url)	no	no	no	full sampling params (temperature/top_p/penalties/logit_bias/logprobs/n/stop)
openai/gpt-4o-mini	no	yes	yes (image_url)	no	no	no	identical request/response shape to gpt-4o
gpt-4o-2024-11-20, gpt-4o-2024-08-06, gpt-4o-2024-05-13	no	yes	yes	no	no	no	dated gpt-4o snapshots; same contract
gpt-4o-mini-2024-07-18	no	yes	yes	no	no	no	dated gpt-4o-mini snapshot
gpt-4o-audio-preview*	no	yes	yes	yes (input_audio)	yes (modalities:[“text”,“audio”] + audio param)	no	audio variant
gpt-4o-mini-audio-preview*	no	yes	yes	yes	yes	no	audio variant
gpt-4o-search-preview*	no	yes	yes	no	no	yes	search variant; emits message.annotations[url_citation]
gpt-4o-mini-search-preview*	no	yes	yes	no	no	yes	search variant
chatgpt-4o-latest	no	yes	yes	no	no	no	rolling alias

All of the above slugs are part of the model enum of the create method. The same endpoint also serves gpt-4.1, gpt-4-turbo, gpt-3.5-turbo, the o1/o3/o4 reasoning models, and the gpt-5.x family.

Critical per-model differences (within openai-chat)

• Sampling params (temperature, top_p, presence_penalty, frequency_penalty, logit_bias, logprobs/top_logprobs, max_tokens, stop, n) are fully supported by gpt-4o / gpt-4o-mini (non-reasoning). max_tokens is Deprecated in favor of max_completion_tokens (both work on gpt-4o; max_tokens is incompatible with o-series reasoning models).
• reasoning_effort, verbosity apply to reasoning / newer models; gpt-4o / gpt-4o-mini ignore them. reasoning_effort defaults: medium for all models before gpt-5.1 (which do not support none); none for gpt-5.1 (gpt-5.1 supports none/low/medium/high); gpt-5-pro defaults to and only supports high; xhigh is supported for all models after gpt-5.1-codex-max.
• stop is NOT supported on the latest reasoning models o3 and o4-mini; works on gpt-4o (up to 4 sequences).
• Audio output (modalities:["text","audio"] + the audio param) is supported ONLY by gpt-4o-audio-preview / gpt-4o-mini-audio-preview. Plain gpt-4o / gpt-4o-mini emit text only (image accepted on input via image_url).
• web_search_options is for *-search-preview variants only.
• Deprecations: user → replaced by safety_identifier + prompt_cache_key; function_call/functions → replaced by tool_choice/tools; seed is Beta and Deprecated; system_fingerprint (response) is Deprecated; max_tokens is Deprecated.
• prompt_cache_retention default depends on org data-retention policy: orgs without ZDR default to 24h, orgs with ZDR default to in_memory (when unspecified). For gpt-5.5, gpt-5.5-pro, and future models only 24h is supported.

Gateway note

In transparent mode the gateway passes the body through unchanged; this contract is what a client may send to openai/gpt-4o[-mini]. Server-behavior fields (store, metadata, service_tier, moderation) affect OpenAI-side behavior and may be irrelevant/inapplicable through the proxy.

Endpoint: POST https://api.openai.com/v1/chat/completions Headers: Authorization: Bearer $OPENAI_API_KEY, Content-Type: application/json Returns: a chat.completion object (or a stream of chat.completion.chunk objects when stream:true). Parameter support varies by model (see per-model differences above).

The Chat Completions resource also exposes stored-completion management methods (not generation): GET /chat/completions (list), GET /chat/completions/{completion_id} (retrieve), POST /chat/completions/{completion_id} (update), DELETE /chat/completions/{completion_id} (delete). These operate on completions stored via store:true and are not part of the create/generate flow. There is no async create/poll variant for this family — generation is synchronous (or streamed via SSE); the async pattern does not apply.

---

1. Request body — shape

{
  "model": "gpt-4o",                 // required
  "messages": [ /* required */ ],    // required, see §2.1
  // --- optional generation / behavior params ---
  "audio": { "format": "...", "voice": "..." },
  "frequency_penalty": 0,
  "logit_bias": { "<token_id>": 0 },
  "logprobs": false,
  "max_completion_tokens": 0,
  "max_tokens": 0,                   // deprecated -> max_completion_tokens
  "metadata": { "<key>": "<value>" },
  "modalities": ["text"],
  "moderation": { "model": "omni-moderation-latest" },
  "n": 1,
  "parallel_tool_calls": true,
  "prediction": { "type": "content", "content": "..." },
  "presence_penalty": 0,
  "prompt_cache_key": "...",
  "prompt_cache_retention": "in_memory" | "24h",
  "reasoning_effort": "...",         // reasoning models
  "response_format": { "type": "text" | "json_object" | "json_schema", ... },
  "safety_identifier": "...",
  "seed": 0,                         // beta/deprecated
  "service_tier": "auto",
  "stop": null,
  "store": false,
  "stream": false,
  "stream_options": { "include_usage": false, "include_obfuscation": true },
  "temperature": 1,
  "tool_choice": "auto",
  "tools": [ /* see §3 */ ],
  "top_logprobs": 0,
  "top_p": 1,
  "user": "...",                     // deprecated
  "verbosity": "medium",             // newer models
  "web_search_options": { ... },     // *-search-preview
  // deprecated
  "function_call": "none" | "auto" | { "name": "..." },
  "functions": [ { "name": "...", "description": "...", "parameters": {...} } ]
}

---

2. Top-level request parameters (ALL)

name	type	required	default	allowed / range	description
messages	array of message objects	yes	—	roles: developer, system, user, assistant, tool, function(deprecated)	List of conversation messages. Supported content types (text/image/audio/file) depend on the model. See §2.1.
model	string (enum)	yes	—	incl. gpt-4o, gpt-4o-2024-11-20, gpt-4o-2024-08-06, gpt-4o-2024-05-13, gpt-4o-mini, gpt-4o-mini-2024-07-18, gpt-4o-audio-preview*, gpt-4o-mini-audio-preview*, gpt-4o-search-preview*, gpt-4o-mini-search-preview*, chatgpt-4o-latest, plus gpt-4.1, o1/o3/o4, gpt-5.x, gpt-4-turbo, gpt-3.5-turbo …	Model ID used to generate the response.
audio	object {format, voice} (ChatCompletionAudioParam)	no	—	see §2.2	Audio-output parameters. Required when audio output is requested via modalities:["audio"].
frequency_penalty	number	no	0	min -2.0, max 2.0	Penalizes tokens by existing frequency in text so far, reducing verbatim repetition.
function_call (Deprecated)	string | object	no	none if no functions, else auto	"none", "auto", or {"name": "my_function"} (ChatCompletionFunctionCallOption {name})	Replaced by tool_choice. Controls which (if any) function is called.
functions (Deprecated)	array of object {name, description, parameters}	no	—	name: a-z A-Z 0-9 _ -, maxLen 64	Replaced by tools. List of functions the model may generate JSON inputs for. Omitting parameters = empty parameter list.
logit_bias	map {token_id: number}	no	null	bias value -100..100	Modifies likelihood of specified tokens. -1..1 nudge; -100/100 effectively ban / force. Effect varies per model.
logprobs	boolean	no	false	true / false	Whether to return log probabilities of output tokens (in content of message).
max_completion_tokens	number	no	—	integer > 0	Upper bound on generated tokens, including visible output AND reasoning tokens.
max_tokens (Deprecated)	number	no	—	integer > 0	Max tokens generated in the chat completion. Replaced by max_completion_tokens; not compatible with o-series models.
metadata	map (Metadata)	no	—	up to 16 pairs; key ≤64 chars, value ≤512 chars	Arbitrary key-value pairs for storing/querying extra info via API/dashboard.
modalities	array of string	no	["text"]	members: "text", "audio"	Output types the model should generate. ["text","audio"] for audio-capable models (e.g. gpt-4o-audio-preview).
moderation	object {model}	no	—	model: e.g. omni-moderation-latest	Configuration for running moderation on the request input and generated output.
n	number	no	1	min 1, max 128	How many chat completion choices to generate per input. Billed across all choices; keep 1 to minimize cost.
parallel_tool_calls	boolean	no	true	true / false	Whether to enable parallel function calling during tool use.
prediction	object (ChatCompletionPredictionContent) {content, type}	no	—	type: "content"	Static predicted output (Predicted Outputs), e.g. content of a regenerated file. See §2.3.
presence_penalty	number	no	0	min -2.0, max 2.0	Penalizes tokens by whether they appear in text so far, increasing likelihood of new topics.
prompt_cache_key	string	no	—	string	Used by OpenAI to cache responses for similar requests (optimizes cache-hit rate). Replaces the user field.
prompt_cache_retention	string (enum)	no	depends on org ZDR policy (24h without ZDR; in_memory with ZDR)	"in_memory", "24h"	Prompt-cache retention policy. 24h = extended caching, prefixes kept active up to 24h. For gpt-5.5 / gpt-5.5-pro / future models only 24h is supported.
reasoning_effort	string (enum)	no	model-dependent (medium pre-5.1; none for gpt-5.1; high for gpt-5-pro)	"none", "minimal", "low", "medium", "high", "xhigh"	Constrains reasoning effort (reasoning models). Less effort = faster / fewer reasoning tokens. (gpt-5.1 supports none/low/medium/high; xhigh only after gpt-5.1-codex-max.)
response_format	object (oneOf)	no	{ "type": "text" }	text | json_object | json_schema	Output format. json_schema = Structured Outputs. See §2.4.
safety_identifier	string	no	—	maxLength 64	Stable per-user ID to help detect policy violations. Recommend a hash of username/email.
seed (Deprecated, Beta)	number	no	—	min -9223372036854775808, max 9223372036854775807	Best-effort deterministic sampling for same seed+params. Determinism not guaranteed; monitor system_fingerprint.
service_tier	string (enum)	no	"auto"	"auto", "default", "flex", "scale", "priority"	Processing type. When set, response echoes the service_tier actually used (may differ from the value sent).
stop	string | array of string	no	null	up to 4 sequences	Stop sequences (excluded from output). NOT supported on latest reasoning models o3 and o4-mini.
store	boolean	no	false	true / false	Whether to store output for model distillation / evals products. Supports text and image inputs (image inputs over 8MB are dropped).
stream	boolean	no	false	true / false	Stream response data as server-sent events as generated.
stream_options	object (ChatCompletionStreamOptions) {include_obfuscation, include_usage}	no	—	only set when stream:true	Streaming options. See §2.5.
temperature	number	no	1	min 0, max 2	Sampling temperature. Higher = more random, lower = more deterministic. Alter this OR top_p, not both.
tool_choice	string | object (oneOf, ChatCompletionToolChoiceOption)	no	none if no tools, else auto	"none", "auto", "required"; allowed_tools; named function/custom	Controls which (if any) tool is called. See §3.2.
tools	array (oneOf: function | custom, ChatCompletionTool)	no	—	function tools and/or custom tools	Tools the model may call. See §3.1.
top_logprobs	number	no	—	integer min 0, max 20	Number of most-likely tokens to return per position (each with logprob). Requires logprobs:true.
top_p	number	no	1	min 0, max 1	Nucleus sampling. 0.1 = only top-10% probability mass considered. Alter this OR temperature.
user (Deprecated)	string	no	—	string	Stable end-user identifier. Being replaced by safety_identifier + prompt_cache_key (use prompt_cache_key for caching).
verbosity	string (enum)	no	medium (newer models)	"low", "medium", "high"	Constrains response verbosity (newer models). Lower = more concise.
web_search_options	object {search_context_size, user_location}	no	—	see §2.6	Web-search tool for use in the response (*-search-preview models).

2.1 messages[] — variants by role

Developer message (ChatCompletionDeveloperMessageParam) {content, role:"developer", name?} — instructions the model follows regardless of the user (replaces system on o1+ models).

• content: string | array of ChatCompletionContentPartText {text, type:"text"} (developer messages support only text parts).
• role: "developer" (required). name: optional string.

System message (ChatCompletionSystemMessageParam) {content, role:"system", name?} — same purpose as developer, for older models.

• content: string | array of text parts (only text). role: "system". name: optional.

User message (ChatCompletionUserMessageParam) {content, role:"user", name?} — end-user content.

• content: string | array of content parts. Part types:
  • ChatCompletionContentPartText {text, type:"text"}
  • ChatCompletionContentPartImage {image_url:{url, detail?}, type:"image_url"} — url: image URL or base64 (format: uri); detail: "auto"(default)|"low"|"high".
  • ChatCompletionContentPartInputAudio {input_audio:{data, format}, type:"input_audio"} — data: base64; format: "wav"|"mp3".
  • FileContentPart {file:{file_data?, file_id?, filename?}, type:"file"} — file_data: base64 file data; file_id: ID of an uploaded file; filename: name when passing as string.
• role: "user". name: optional.

Assistant message (ChatCompletionAssistantMessageParam) {role:"assistant", audio?, content?, function_call?(deprecated), name?, refusal?, tool_calls?}

• audio: optional {id} — reference to a previous audio response from the model.
• content: optional string | array of (ChatCompletionContentPartText {text, type:"text"} | ChatCompletionContentPartRefusal {refusal, type:"refusal"}). Can be one or more text, or exactly one refusal. Required unless tool_calls or function_call is specified.
• function_call (deprecated): {arguments, name}.
• refusal: optional string. name: optional string.
• tool_calls: optional array of:
  • function tool call: {id, function:{arguments, name}, type:"function"}
  • custom tool call: {id, custom:{input, name}, type:"custom"}

Tool message (ChatCompletionToolMessageParam) {content, role:"tool", tool_call_id}

• content: string | array of text parts. role: "tool". tool_call_id: string (ID of the tool call this responds to).

Function message (Deprecated) (ChatCompletionFunctionMessageParam) {content, name, role:"function"}.

2.2 audio (ChatCompletionAudioParam — request, output config)

field	type	required	allowed
format	string (enum)	yes (within object)	"wav", "aac", "mp3", "flac", "opus", "pcm16"
voice	string | enum | object {id}	yes (within object)	built-in voices: alloy, ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer, verse, marin, cedar; OR a custom voice object, e.g. { "id": "voice_1234" }.

2.3 prediction (ChatCompletionPredictionContent)

field	type	required	allowed
content	string | array of ChatCompletionContentPartText {text, type:"text"}	yes	The content to match when generating; if generated tokens match, the response can be returned faster.
type	string	yes	"content" (currently always content)

2.4 response_format (oneOf)

• { "type": "text" } (ResponseFormatText) — default, plain text response.

• { "type": "json_object" } (ResponseFormatJSONObject) — older JSON mode; requires a system/user instruction to produce JSON.

• { "type": "json_schema", "json_schema": {...} } (ResponseFormatJSONSchema) — Structured Outputs:

  field	type	required	allowed
  name	string	yes	a-z A-Z 0-9 _ -, maxLen 64
  description	string	no	—
  schema	map (JSON Schema object)	no	JSON Schema object
  strict	boolean	no	true = strict schema adherence; only a subset of JSON Schema supported when strict is true
  type	string	yes	"json_schema"

2.5 stream_options (ChatCompletionStreamOptions)

field	type	required	default	description
include_obfuscation	boolean	no	true (included by default)	Adds random chars to an obfuscation field on streaming delta events to normalize payload sizes (side-channel mitigation). Set false to optimize bandwidth on trusted links.
include_usage	boolean	no	false	If set, an extra chunk is streamed before data: [DONE] whose usage field holds whole-request token stats and choices is []. All other chunks carry usage:null. If the stream is interrupted, the final usage chunk may not arrive.

2.6 web_search_options

field	type	required	default	allowed
search_context_size	string (enum)	no	medium	"low", "medium", "high"
user_location	object {approximate, type:"approximate"}	no	—	approximate {city?, country?, region?, timezone?}; country = two-letter ISO; timezone = IANA (e.g. America/Los_Angeles); city/region free text; type always "approximate"

---

3. tools / tool_choice

3.1 tools[] (oneOf, ChatCompletionTool)

Function tool (ChatCompletionFunctionTool) {type:"function", function:{name, description?, parameters?, strict?}}

field	type	required	allowed
function.name	string	yes	a-z A-Z 0-9 _ -, maxLen 64
function.description	string	no	—
function.parameters	object (FunctionParameters / JSON Schema)	no	omitting = empty parameter list
function.strict	boolean	no	true = strict schema adherence (Structured Outputs); only a subset of JSON Schema supported when strict
type	string	yes	"function" (currently only function supported as a tool type here)

Custom tool (ChatCompletionCustomTool) {type:"custom", custom:{name, description?, format?}}

• custom.name: string (required) — identifies the tool in tool calls.
• custom.description: optional string.
• custom.format (oneOf, default = unconstrained text):
  • TextFormat {type:"text"} — unconstrained free-form text.
  • GrammarFormat {type:"grammar", grammar:{definition, syntax}} — definition: grammar definition string; syntax: "lark" | "regex".

3.2 tool_choice (oneOf, ChatCompletionToolChoiceOption)

• ToolChoiceMode = "none" | "auto" | "required". (none = no tool, generate a message; auto = model picks; required = must call ≥1 tool.)
• Allowed tools (ChatCompletionAllowedToolChoice) {type:"allowed_tools", allowed_tools:{mode, tools}} — mode: "auto"|"required"; tools: array of tool definition maps, e.g. [{ "type":"function", "function":{ "name":"get_weather" } }].
• Named function (ChatCompletionNamedToolChoice) {type:"function", function:{name}}.
• Named custom (ChatCompletionNamedToolChoiceCustom) {type:"custom", custom:{name}}.

Default: none when no tools present, auto when tools present.

---

4. Response — chat.completion object

{
  "id": "string",
  "object": "chat.completion",
  "created": 1741569952,            // unix time
  "model": "gpt-4o-...",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "string|null",
        "refusal": "string|null",
        "annotations": [ { "type": "url_citation", "url_citation": {...} } ],
        "audio": { "id", "data", "expires_at", "transcript" },   // if audio output
        "tool_calls": [ ... ],
        "function_call": { "arguments", "name" }                 // deprecated
      },
      "logprobs": { "content": [...], "refusal": [...] } | null,
      "finish_reason": "stop"
    }
  ],
  "usage": { ... },
  "service_tier": "default",
  "system_fingerprint": "string",   // deprecated
  "moderation": { "input": {...}, "output": {...} }  // if moderated completions requested
}

4.1 Response fields

field	type	description
id	string	Unique identifier for the chat completion.
choices	array	List of choices (>1 if n>1).
choices[].finish_reason	enum	"stop", "length", "tool_calls", "content_filter", "function_call"(deprecated).
choices[].index	number	Index of the choice in the list.
choices[].logprobs	object | null	{content[], refusal[]}; each item is a ChatCompletionTokenLogprob {token, bytes[](nullable), logprob, top_logprobs[]}; top_logprobs[] entries are {token, bytes[], logprob}; logprob = -9999.0 if outside top-20.
choices[].message.content	string	Message contents.
choices[].message.refusal	string	Refusal message generated by the model.
choices[].message.role	"assistant"	Role of the author.
choices[].message.annotations	array	{type:"url_citation", url_citation:{end_index, start_index, title, url}} (present when using web search).
choices[].message.audio	object	{id, data(base64), expires_at(unixtime), transcript} (if audio output modality requested).
choices[].message.function_call (deprecated)	object	{arguments, name}.
choices[].message.tool_calls	array	function: ChatCompletionMessageFunctionToolCall {id, function:{arguments, name}, type:"function"}; custom: ChatCompletionMessageCustomToolCall {id, custom:{input, name}, type:"custom"}.
created	number	Unix timestamp (seconds) of creation.
model	string	Model used for the completion.
object	"chat.completion"	Object type (always chat.completion).
service_tier	enum	Tier actually used: auto/default/flex/scale/priority.
system_fingerprint (Deprecated)	string	Backend-configuration fingerprint (use with seed to detect backend changes).
moderation	object	{input, output} (present if moderated completions were requested). Each side is ModerationResults {model, results[], type:"moderation_results"} OR Error {code, message, type:"error"}. Each entry in results[] is {categories(map[bool]), category_applied_input_types(map[array of "text"/"image"]), category_scores(map[number]), flagged(bool), model(string), type:"moderation_result"}.
usage	object (CompletionUsage)	Token usage statistics; see §4.2.

4.2 usage (CompletionUsage)

field	type	description
completion_tokens	number	Tokens in the generated completion.
prompt_tokens	number	Tokens in the prompt.
total_tokens	number	Total = prompt + completion.
completion_tokens_details	object	{accepted_prediction_tokens?, audio_tokens?, reasoning_tokens?, rejected_prediction_tokens?}. accepted_prediction_tokens / rejected_prediction_tokens = Predicted-Outputs hits/misses (rejected still billed); reasoning_tokens = tokens spent on reasoning; audio_tokens = audio tokens generated by the model.
prompt_tokens_details	object	{audio_tokens?, cached_tokens?}. audio_tokens = audio input tokens in the prompt; cached_tokens = cached tokens in the prompt.

4.3 Streaming chunk (chat.completion.chunk)

When stream:true, the response is a stream of chat.completion.chunk objects carrying choices[].delta (incremental) instead of choices[].message. With stream_options.include_usage:true, a final chunk before data: [DONE] carries the whole-request usage and choices = []; all other chunks carry usage: null.

---

5. MINIMAL example

Request

curl https://api.openai.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
    "model": "gpt-4o",
    "messages": [
      { "role": "developer", "content": "You are a helpful assistant." },
      { "role": "user", "content": "Hello!" }
    ]
  }'

Response

{
  "id": "chatcmpl-B9MBs8CjcvOU2jLn4n570S5qMJKcT",
  "object": "chat.completion",
  "created": 1741569952,
  "model": "gpt-4o",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Hello! How can I assist you today?",
        "refusal": null,
        "annotations": []
      },
      "logprobs": null,
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 19,
    "completion_tokens": 10,
    "total_tokens": 29,
    "prompt_tokens_details": { "cached_tokens": 0, "audio_tokens": 0 },
    "completion_tokens_details": {
      "reasoning_tokens": 0, "audio_tokens": 0,
      "accepted_prediction_tokens": 0, "rejected_prediction_tokens": 0
    }
  },
  "service_tier": "default"
}

Streaming (stream:true) — stream of chunks:

{"id":"chatcmpl-123","object":"chat.completion.chunk","created":1694268190,"model":"gpt-4o-mini","system_fingerprint":"fp_44709d6fcb","choices":[{"index":0,"delta":{"role":"assistant","content":""},"logprobs":null,"finish_reason":null}]}
{"id":"chatcmpl-123","object":"chat.completion.chunk","created":1694268190,"model":"gpt-4o-mini","system_fingerprint":"fp_44709d6fcb","choices":[{"index":0,"delta":{"content":"Hello"},"logprobs":null,"finish_reason":null}]}
{"id":"chatcmpl-123","object":"chat.completion.chunk","created":1694268190,"model":"gpt-4o-mini","system_fingerprint":"fp_44709d6fcb","choices":[{"index":0,"delta":{},"logprobs":null,"finish_reason":"stop"}]}

---

6. FULL examples (vision + tools + structured-style options + audio)

Vision (image input) — request

curl https://api.openai.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
    "model": "gpt-4o",
    "messages": [
      {
        "role": "user",
        "content": [
          { "type": "text", "text": "What is in this image?" },
          { "type": "image_url",
            "image_url": { "url": "https://.../boardwalk.jpg", "detail": "auto" } }
        ]
      }
    ],
    "max_tokens": 300
  }'

Function calling — request

curl https://api.openai.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
    "model": "gpt-4o",
    "messages": [
      { "role": "user", "content": "What is the weather like in Boston today?" }
    ],
    "tools": [
      {
        "type": "function",
        "function": {
          "name": "get_current_weather",
          "description": "Get the current weather in a given location",
          "parameters": {
            "type": "object",
            "properties": {
              "location": { "type": "string", "description": "The city and state, e.g. San Francisco, CA" },
              "unit": { "type": "string", "enum": ["celsius", "fahrenheit"] }
            },
            "required": ["location"]
          },
          "strict": true
        }
      }
    ],
    "tool_choice": "auto",
    "parallel_tool_calls": true,
    "temperature": 0.7,
    "top_p": 1,
    "frequency_penalty": 0,
    "presence_penalty": 0,
    "n": 1,
    "stop": null,
    "max_completion_tokens": 512,
    "response_format": { "type": "text" },
    "logprobs": true,
    "top_logprobs": 5,
    "seed": 42,
    "store": false,
    "metadata": { "session": "abc" },
    "service_tier": "auto",
    "stream": false,
    "prompt_cache_key": "user-123"
  }'

Function calling — response

{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "created": 1699896916,
  "model": "gpt-4o-mini",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": null,
        "tool_calls": [
          {
            "id": "call_abc123",
            "type": "function",
            "function": {
              "name": "get_current_weather",
              "arguments": "{\n\"location\": \"Boston, MA\"\n}"
            }
          }
        ]
      },
      "logprobs": null,
      "finish_reason": "tool_calls"
    }
  ],
  "usage": {
    "prompt_tokens": 82,
    "completion_tokens": 17,
    "total_tokens": 99,
    "completion_tokens_details": {
      "reasoning_tokens": 0,
      "accepted_prediction_tokens": 0,
      "rejected_prediction_tokens": 0
    }
  }
}

Structured Outputs — response_format with json_schema (request fragment)

"response_format": {
  "type": "json_schema",
  "json_schema": {
    "name": "weather",
    "description": "structured weather",
    "strict": true,
    "schema": {
      "type": "object",
      "properties": { "temp_c": { "type": "number" } },
      "required": ["temp_c"],
      "additionalProperties": false
    }
  }
}

Audio output (audio variant model) — request fragment

{
  "model": "gpt-4o-audio-preview",
  "modalities": ["text", "audio"],
  "audio": { "voice": "alloy", "format": "wav" },
  "messages": [ { "role": "user", "content": "Is a golden retriever a good family dog?" } ]
}

Audio input (audio variant model) — user message content part

{
  "role": "user",
  "content": [
    { "type": "text", "text": "Transcribe and answer this." },
    { "type": "input_audio", "input_audio": { "data": "<base64>", "format": "mp3" } }
  ]
}