GPT-5: Key characteristics, pricing and system card
17 days ago
- #AI Models
- #OpenAI
- #GPT-5
- GPT-5 is a hybrid model switching between different models based on conversation type and complexity.
- GPT-5 in the API is available as three models: regular, mini, and nano, each with four reasoning levels.
- The model has an input limit of 272,000 tokens and an output limit of 128,000 tokens.
- GPT-5 is priced competitively: $1.25/million for input and $10/million for output.
- The model family replaces most of OpenAI's previous models, with GPT-5 Pro available only in the $200/month tier.
- GPT-5 focuses on reducing hallucinations, improving instruction following, and minimizing sycophancy.
- Safe-completions approach moderates answers to avoid harmful content instead of outright refusals.
- Prompt injection remains a challenge with a 56.8% attack success rate for GPT-5.
- The API includes reasoning traces and a minimal reasoning effort option for faster responses.
- GPT-5 performs well in generating SVGs, as demonstrated by a pelican riding a bicycle.