No longer supported

This model is retired (released January 2022) and is no longer actively maintained or recommended by OpenAI. Consider using their current flagship model instead.

retiredlegacyllm Proprietary

InstructGPT davinci

by OpenAI· Released January 2022· Cutoff June 2021

InstructGPT davinci is a fine-tuned version of GPT-3 that follows instructions more reliably and generates safer outputs. It was one of the first models to use reinforcement learning from human feedback (RLHF). It has been superseded by newer models like GPT-3.5 Turbo and GPT-4.

Official Site 📄 Research Paper

Input cost

$0.0200 per 1K tokens

Output cost

$0.0200 per 1K tokens

Context window

2049 tokens

Max output

2049 tokens

Modalities

text

Parameters

175B

License

proprietary

Capabilities

Text GenerationInstruction FollowingZero-shot LearningFew-shot LearningCode Generation

Best For

General-purpose text generation and instruction following tasks where high-quality outputs are needed.

Strengths

Excellent at following detailed instructions
Produces coherent and contextually relevant text
Good zero-shot and few-shot performance

Limitations

Small context window (2049 tokens)
No longer supported or available
Higher latency compared to newer models
May generate biased or toxic content if not carefully prompted

Use Cases

Content creation and copywriting

Question answering

Summarization

Translation

Code generation and explanation

Creative writing

Chatbots and virtual assistants

Improvements Over Previous Model

First model to use RLHF for instruction following
Significantly better at following user intent compared to GPT-3 base
Reduced toxic and biased outputs compared to GPT-3
Introduced InstructGPT family, later evolved into GPT-3.5 and GPT-4

Back to all models

No longer supported

This model is retired (released January 2022) and is no longer actively maintained or recommended by OpenAI. Consider using their current flagship model instead.

retiredlegacyllm Proprietary

InstructGPT davinci

by OpenAI· Released January 2022· Cutoff June 2021

Official Site 📄 Research Paper

Input cost

$0.0200 per 1K tokens

Output cost

$0.0200 per 1K tokens

Context window

2049 tokens

Max output

2049 tokens

Modalities

text

Parameters

175B

License

proprietary

Capabilities

Text GenerationInstruction FollowingZero-shot LearningFew-shot LearningCode Generation

Best For

General-purpose text generation and instruction following tasks where high-quality outputs are needed.

Strengths

Excellent at following detailed instructions
Produces coherent and contextually relevant text
Good zero-shot and few-shot performance

Limitations

Small context window (2049 tokens)
No longer supported or available
Higher latency compared to newer models
May generate biased or toxic content if not carefully prompted

Use Cases

Content creation and copywriting

Question answering

Summarization

Translation

Code generation and explanation

Creative writing

Chatbots and virtual assistants

Improvements Over Previous Model

First model to use RLHF for instruction following
Significantly better at following user intent compared to GPT-3 base
Reduced toxic and biased outputs compared to GPT-3
Introduced InstructGPT family, later evolved into GPT-3.5 and GPT-4

Back to all models