OpenAI logo

GPT-5 nano

Multimodal
OpenAI

GPT-5 nano is our fastest and cheapest GPT-5 variant. It excels at summarization and classification tasks with moderate reasoning capabilities and very high speed.

Key Specifications

Parameters
-
Context
400.0K
Release Date
August 7, 2025
Average Score
50.1%

Timeline

Key dates in the model's history
Announcement
August 7, 2025
Today
March 26, 2026

Technical Specifications

Parameters
-
Training Tokens
-
Knowledge Cutoff
May 30, 2024
Family
-
Capabilities
MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)
$0.05
Output (per 1M tokens)
$0.40
Max Input Tokens
400.0K
Max Output Tokens
128.0K
Supported Features
Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Reasoning

Logical reasoning and analysis
GPQA
GPT-5 nano - Diamond thinking without tools This method uses mode thinking Diamond for solutions complex tasks without use tools. Diamond thinking represents itself structured approach to and solving problems, where model: 1. task on parts 2. various solutions 3. works over each step 4. results, in order to errors GPT-5 nano demonstrates abilities reasoning in this mode, especially for mathematical tasks and logical puzzles. Model can its approach, errors and its thinking by In tests on complex mathematical tasks from AIME and FrontierMath, this method improvement by comparison with approach without thinking. Especially efficient for tasks, requiring reasoning and analysisSelf-reported
71.2%

Other Tests

Specialized benchmarks
AIME 2025
GPT-5 nano with included mode thinking (without tools) - mathematicsSelf-reported
85.2%
FrontierMath
GPT-5 nano with included mode thinking (only with Python-tool) - FrontierMath levels 1-3 mathematics expert level.Self-reported
9.6%
Humanity's Last Exam
GPT-5 nano with mode thinking (without tools) - set questions expert level by various subjectsSelf-reported
8.7%
HMMT 2025
GPT-5 nano with included mode thinking (without tools) - Harvard-MIT Mathematics Tournament.Self-reported
75.6%

License & Metadata

License
proprietary
Announcement Date
August 7, 2025
Last Updated
July 24, 2025

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.