What is GPT-3.5 Turbo?

GPT-3.5 Turbo is a low-cost model optimized for chat, classification, and extraction tasks, with a mature fine-tuning ecosystem.

How good is GPT-3.5 Turbo?

It's reliable for simple tasks, and fine-tuned versions can match GPT-4 quality on narrow tasks.

What can GPT-3.5 Turbo do?

It excels at simple chat, classification, extraction, and fine-tuning use cases at low cost.

Should I use GPT-3.5 Turbo or GPT-4o-mini?

GPT-4o-mini is cheaper and more capable for new development. Use GPT-3.5 Turbo for legacy apps or fine-tuning.

How much does GPT-3.5 Turbo cost?

GPT-3.5 Turbo pricing is $0.50 per 1M input tokens and $1.50 per 1M output tokens.

LangFast

Beta

GPT-3.5 Turbo

Low-cost model for simple chat, classification, and extraction.

Try in Playground

Community Sentiment

Mostly Positive

Based on Reddit reviews

Community Verdict

Best for Fine-Tuning

Based on Reddit reviews

Input Modalities

Text

Output Modalities

Text

Price / 1M tokens

$0.5$1.5

InputOutput

Best For

Fine-tuning use cases

Long-term support for legacy applications

Budget-conscious chat applications

High-volume, low-complexity tasks

Avoid For

All new development (use GPT-4o-mini instead)

Complex reasoning tasks (use GPT-5 instead)

Vision/multimodal tasks (not supported)

Audio processing (not supported)

16,385 context window

4,096 max output tokens

Sep 1, 2021 knowledge cutoff

Reasoning not supported

Parameters

While OpenAI documents a unified parameter set for the Chat Completions and Responses APIs, each model supports only a limited subset of model-specific parameters and values. This table lists the supported model-specific GPT-3.5 Turbo API parameters and allowed values.

Parameter	Description	Path	Supported Values
Model	Selects the model version for the request	model	gpt-3.5-turbogpt-3.5-turbo-0125
Message roles	Defines the role of the message in the input	messages[].role	developersystemuserassistant
Max output tokens	Max output tokens the model may generate	max_completion_tokens	16 .. 4,096
Output format	Specifies the output format, including structured JSON schemas	response_format	textjson_object
Temperature	Controls how random or deterministic the output is	temperature	0 .. 2
Top P	Controls how diverse the output tokens are	top_p	0 .. 1
Presence penalty	Encourages the model to introduce new topics	presence_penalty	-2 .. 2
Frequency penalty	Reduces repetition of the same words or phrases	frequency_penalty	-2 .. 2
Reasoning effort	Controls the depth of internal reasoning used by the model	reasoning_effort	Not supported
Reasoning summary	Controls whether the model produces a concise or detailed reasoning summary	reasoning_summary	Not supported
Verbosity	Controls how brief or detailed the generated response is	verbosity	Not supported

Pricing

GPT-3.5 Turbo API pricing is based on token usage for input and output. Prices are listed per 1M tokens, with lower rates for cached input. Tool-specific features may add per-call fees.

Text tokens

Per 1M tokens

Input$0.50

Cached inputN/A

Output$1.50

Example costs (GPT-3.5 Turbo)

TASK

APPROX COST

Simple chat (1k tokens)

~$0.0005

Classification batch (10k items)

~$0.05

Text extraction

~$0.01–$0.05

Modalities

What the model can accept and produce

TextInput and output

ImagesNot supported

AudioNot supported

VideoNot supported

Features

Platform-level capabilities

StreamingNot supported

Function callingNot supported

Structured outputsNot supported

Fine-tuningSupported

DistillationNot supported

Predicted outputsNot supported

Tools

Tools supported by this model when using the Responses API.

Web searchNot supported

File searchSupported

Image generationNot supported

Code interpreterSupported

MCPSupported

Computer useNot supported

Snapshots

GPT-3.5 Turbo model snapshots ensure stable behavior by locking a specific version. See all available snapshots and aliases below.

gpt-3.5-turbo

↪ gpt-3.5-turbo-0125

gpt-3.5-turbo-0125

gpt-3.5-turbo-1106Deprecated

gpt-3.5-turbo-instructDeprecated

FAQs

Ship prompts that pass the tests

Don't wait until they break in production

Get started – it's free