OpenAI's versatile, high-intelligence flagship model. Accepts text and image inputs, produces text outputs including structured outputs. Best model for most tasks outside reasoning-heavy use cases.
GPT-4o ('omni'), released May 13, 2024, is OpenAI's flagship multimodal model that accepts text, image, and audio inputs and produces text and audio outputs. It achieved GPT-4 Turbo-class intelligence at half the price and with near-real-time voice conversation latency — a step change in accessibility and cost efficiency for production AI workloads. GPT-4o scored 88.7 % on MMLU, >90 % on HumanEval, and set new standards for vision understanding across image, PDF, and diagram inputs.
From a TheRouter operator perspective, GPT-4o occupies a unique position: it is the most broadly capable single model for mixed-modality workloads that don't require chain-of-thought reasoning (covered by the o-series) or the extreme cost efficiency of GPT-4o Mini. As of May 2026, GPT-4o remains widely listed on TheRouter and is still active in the OpenAI API, though it was retired from ChatGPT on February 13, 2026 in favour of GPT-5.2. Its successor — GPT-4.1 — matches or exceeds GPT-4o's intelligence evals while reducing latency by nearly half and cost by 83 %, making GPT-4o primarily relevant for workflows that depend on its specific prompt-compatibility profile or voice/multimodal pipeline stability.
| Type | Rate |
|---|---|
| Input | $2.70 / 1M tokens |
| Output | $10.80 / 1M tokens |
| Release date | 2024-05-13openai.com ↗ | verified |
| Latest snapshot | gpt-4o-2024-11-20 (Nov '24); earlier snapshot gpt-4o-2024-08-06 (Aug '24); original gpt-4o-2024-05-13developers.openai.com ↗ | verified |
| Training cutoff | Not publicly disclosed (estimated October 2023) | single source |
| Input modalities | Text, image (including PDF and diagrams), audioopenai.com ↗ | verified |
| Output modalities | Text (default), audio (realtime API), structured JSON (response_format)openai.com ↗ | verified |
| Structured output support | Yes — native response_format (json_schema, json_object)platform.openai.com ↗ | verified |
| Max output tokens | 16 384 (standard); 4 096 (audio mode via realtime API)developers.openai.com ↗ | verified |
| Deployment | OpenAI API, Azure OpenAI, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundryopenai.com ↗ | verified |
| License | OpenAI Terms of Service (proprietary, API-only) | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
MMLU OpenAI's reported score at launch — near-expert performance across 57 subjects. | 88.7% | openai.com ↗ | |
HumanEval GPT-4o achieves near-expert level Python code generation across a range of prompt complexities. | ~90.2% | openai.com ↗ | |
MMMU Massive Multi-discipline Multimodal Understanding benchmark — measures vision-language reasoning across 6 disciplines. | 69.1% | openai.com ↗ | |
MATH Reported in OpenAI's system card. Zero-shot CoT prompting. | 76.6% | openai.com ↗ | |
GSM8K Grade-school math word problems — eight-shot CoT. GPT-4o is competitive with GPT-4 Turbo. | 90.5% | openai.com ↗ |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "openai/gpt-4o",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Standard chat or text-in/text-out completion via TheRouter's OpenAI-compatible endpoint.
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"openai/gpt-4o","messages":[{"role":"user","content":"Explain what GPT-4o is in one sentence."}]}'OpenAI retired GPT-4o from ChatGPT on February 13, 2026 alongside GPT-4.1, GPT-4.1 mini, and o4-mini in favour of GPT-5.2. The model remains available in the OpenAI API and through TheRouter with no announced API deprecation.
OpenAI released GPT-4.1, matching or exceeding GPT-4o in intelligence evals while reducing latency by nearly half and cost by 83%. The GPT-4.1 family includes GPT-4.1 mini and GPT-4.1 nano as smaller, cheaper alternatives.
OpenAI released the gpt-4o-2024-11-20 snapshot with improved performance and lower latency. This is the recommended snapshot for new integrations as of May 2026.
Yes. GPT-4o was retired from ChatGPT on February 13, 2026, but remains fully available in the OpenAI API and through TheRouter with no announced API deprecation date. The gpt-4o-2024-11-20 snapshot is the recommended version.
GPT-4o is the flagship multimodal model with full vision, audio, and structured output capabilities at $2.50/$10 per million tokens. GPT-4o Mini is a lightweight, text-focused variant at $0.15/$0.60 per million tokens (17x cheaper) with reduced intelligence but strong performance for simple tasks. Both support 128K context.
GPT-4o natively accepts audio input and can output audio in near real-time (as low as 232 ms voice-to-voice latency). However, audio input/output requires OpenAI's Realtime API endpoints (/v1/realtime) rather than the standard /v1/chat/completions endpoint. On TheRouter, use the openai/realtime-preview or openai/gpt-4o-audio-preview model aliases for voice applications.
No. GPT-4o is a general-purpose multimodal model without explicit chain-of-thought reasoning. For complex multi-step reasoning, coding, or math tasks, route to openai/o3, o4-mini, or o3-pro which are designed for extended thinking.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | openai.com ↗ | 2026-05-28 | verified |
| Latest snapshot | developers.openai.com ↗ | 2026-05-28 | verified |
| Training cutoff | — | — | single source |
| Input modalities | openai.com ↗ | 2026-05-28 | verified |
| Output modalities | openai.com ↗ | 2026-05-28 | verified |
| Structured output support | platform.openai.com ↗ | 2026-05-28 | verified |
| Max output tokens | developers.openai.com ↗ | 2026-05-28 | verified |
| Deployment | openai.com ↗ | 2026-05-28 | verified |
| License | — | — | verified |
| MMLU | openai.com ↗ | 2026-05-28 | verified |
| HumanEval | openai.com ↗ | 2026-05-28 | verified |
| MMMU | openai.com ↗ | 2026-05-28 | verified |
| MATH | openai.com ↗ | 2026-05-28 | verified |
| GSM8K | openai.com ↗ | 2026-05-28 | verified |
| GPT-4o retired from ChatGPT | openai.com ↗ | 2026-05-28 | verified |
| GPT-4.1 announced as GPT-4o successor | openai.com ↗ | 2026-05-28 | verified |
| GPT-4o November 2024 snapshot released | openai.com ↗ | 2026-05-28 | verified |
| Is GPT-4o still available in the API? | openai.com ↗ | 2026-05-28 | to verify |
| What is the difference between GPT-4o and GPT-4o Mini? | openai.com ↗ | 2026-05-28 | to verify |
| Can GPT-4o process audio inputs? | openai.com ↗ | 2026-05-28 | to verify |
| Is GPT-4o a reasoning model? | openai.com ↗ | 2026-05-28 | to verify |