GPT-4o
GPT-4o is an OpenAI multimodal model that accepts text and image inputs and produces text. It supports streaming, function calling, Structured Outputs, fine-tuning, and predicted outputs for vision-heavy assistants and structured extraction workflows.
Newer version: GPT-5.4
Provider
OpenAI
Model family
OpenAI GPT
Multimodal LLM
Cost tier
Flagship
Status
Legacy
Release May 13, 2024
Why teams choose it
Azure OpenAI and direct OpenAI APIs differ slightly in SKU names—pin deployment names in config.
Multimodal limits (image count, resolution) change
validate against the current model card before UX sign-off.
Tradeoffs to know
- Pricing and rate limits are tier-dependent—budget for burst traffic.
- Policy and safety defaults differ between consumer ChatGPT and API products.
Technical specs
- Inputs
- text, image
- Outputs
- text
- Capabilities
- tool use, vision, structured outputs, function calling, streaming, fine-tuning
- License
- Proprietary API
Benchmarks
- Mmlu
- 88.7
- Humaneval
- 90.2
GPT-4o features
GPT-4o accepts text and image inputs and produces text outputs. OpenAI documents a 128,000-token context window and a maximum output of 16,384 tokens for the current GPT-4o API model.
- Streaming, function calling, Structured Outputs, fine-tuning, and predicted outputs are supported.
- Image input is supported; the base gpt-4o model page lists text output and does not list native audio or video input for this model ID.
- Use a dated snapshot when behavior stability matters, then regression-test before changing snapshots.
Sources: OpenAI GPT-4o model reference
GPT-4o capabilities and production fit
GPT-4o remains useful for multimodal extraction, visual question answering, tool-enabled assistants, structured data generation, and applications that already have a measured GPT-4o quality baseline.
- Evaluate vision quality with the same image types, resolutions, and failure cases used in production.
- Validate function arguments and Structured Outputs in application code before executing side effects.
- Measure latency, token use, refusal behavior, and task success instead of relying on a generic capability label.
Sources: OpenAI GPT-4o model reference
GPT-4o deployment options
OpenAI lists GPT-4o for the Responses API, Chat Completions API, Batch API, and fine-tuning. Choose the endpoint that matches the application lifecycle, and keep the model ID, timeout policy, retries, output validation, and evaluation set explicit in configuration.
- Use Responses for new stateful or tool-using integrations when its features match the workload.
- Keep Chat Completions where an existing integration is stable and a migration has not been justified by measured benefit.
- Treat ChatGPT aliases and dedicated realtime, audio, transcription, or search models as separate products rather than assuming the base gpt-4o ID exposes every GPT-4o-branded capability.
OpenAI GPT family lineup
Current models
Related models
GPT-4o FAQ
What are GPT-4o's main capabilities?
GPT-4o is an OpenAI multimodal model used for chat, tool calling, vision, structured extraction, and interactive product workflows. Confirm modality support, context limits, pricing, and availability for the exact OpenAI or Azure API surface you plan to use.
Explore next
Models, tools, and comparisons that connect to this reference.