# GPT-4o ![Organization](https://img.shields.io/badge/Org-OpenAI-412991) ![Context](https://img.shields.io/badge/Context-128K%20tokens-1f6feb) ![Max output](https://img.shields.io/badge/Max%20output-16%2C384%20tokens-1f6feb) ![Pricing](https://img.shields.io/badge/Price-%242.50%20%2F%20%2410%20per%201M-2da44e) ![License](https://img.shields.io/badge/License-Proprietary-d29922) ![Modality](https://img.shields.io/badge/Modality-Text%20%2B%20Vision%20%2B%20Audio-8957e5) ![Released](https://img.shields.io/badge/Released-May%202024-555) > **Quick answer:** GPT-4o ("o" for "omni") is OpenAI's flagship multimodal model, released May 13, 2024. It processes text, vision, and audio in a single end-to-end neural network, runs at twice the speed of GPT-4 Turbo at half the price, and delivers 128K-token context with a 16,384-token max output. At $2.50/$10 per million tokens, it is one of the most cost-efficient frontier models for production API use. ## At a Glance **Where GPT-4o leads** - True end-to-end multimodal — text, vision, and audio handled natively in one model (no pipeline stitching) - Competitive pricing: $2.50/$10 per 1M tokens, 50% cheaper than GPT-4 Turbo at release - Strong multilingual performance with a new tokenizer delivering up to 4× fewer tokens on non-Latin scripts - Available on the free ChatGPT tier, maximising developer familiarity and ecosystem tooling **Where it lags** - 128K context window, smaller than Claude 3.5 Sonnet (200K) or Gemini 1.5 Pro (2M) at similar price points - Superseded for complex reasoning tasks by GPT-4.1, o-series reasoning models, and eventually GPT-5 - Knowledge cutoff October 2023; events after that require retrieval augmentation **Best for:** production workloads needing reliable multimodal reasoning — vision analysis, voice interfaces, and text tasks — at a price point that scales to high volume. ## What GPT-4o Is GPT-4o is OpenAI's first model to combine text, vision, and audio in a single end-to-end neural network trained across all three modalities simultaneously. Earlier voice pipelines stitched three separate models together (speech-to-text → LLM → text-to-speech), losing tone, emotion, and speaker nuance at each handoff. GPT-4o eliminates those handoffs: it reads audio directly, interprets it alongside visual context, and produces native audio output — with response latency as low as 232ms, close to human conversational timing. The model matches GPT-4 Turbo on English text and code benchmarks while running twice as fast and costing 50% less per token at launch. A redesigned tokenizer reduces token counts on non-Latin script languages by up to 4×, materially cutting API costs for multilingual applications. The November 2024 update (`gpt-4o-2024-11-20`) improved structured output reliability and tool-use performance, making it more suitable as an agent backbone. From a Benchgen perspective, GPT-4o is a high-volume production model: broad ecosystem support (available in ChatGPT free, Plus, Teams, Enterprise, API, Azure OpenAI, and more), well-understood failure modes, and strong community tooling. Its agent trajectory data is among the most widely collected in the industry, which matters when using Benchgen to analyse baseline performance or set score thresholds for comparison. ## Specifications | Field | Value | |---|---| | Organization | OpenAI | | Model type | Omni multimodal model (text, vision, audio) | | API identifier | `gpt-4o-2024-11-20` (latest snapshot) | | Access | Closed, API (OpenAI API, Azure OpenAI Service, ChatGPT) | | Context window | 128,000 tokens | | Max output | 16,384 tokens | | Architecture | Dense transformer with native audio/vision encoders (details undisclosed) | | License | Proprietary | | Release date | May 13, 2024 (original); November 20, 2024 (latest snapshot) | | Knowledge cutoff | October 2023 | | Modality | Multimodal (text, vision, audio) | ## Pricing | | Input (per 1M tokens) | Output (per 1M tokens) | |---|---|---| | OpenAI | $2.50 | $10.00 | Cached input tokens are billed at $1.25/1M; batch API processing reduces costs by 50%. Source: [OpenAI pricing page](https://openai.com/api/pricing/). ## Context Window GPT-4o has a **128,000-token context window** — roughly **200 pages** of text in a single request — with a maximum output of **16,384 tokens**. The window is sufficient for large document analysis, multi-file code review, and conversational memory across long sessions. For tasks requiring longer context (entire codebases or book-length documents), Claude 3.5 Sonnet (200K) or Gemini 1.5 Pro (2M) offer more headroom at a similar price tier. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | GPQA Diamond | 53.6% | [OpenAI GPT-4o system card](https://openai.com/index/gpt-4o-system-card/) | 2024-05 | | MMLU | 88.7% | [OpenAI GPT-4o system card](https://openai.com/index/gpt-4o-system-card/) | 2024-05 | | HumanEval | 90.2% | [OpenAI GPT-4o system card](https://openai.com/index/gpt-4o-system-card/) | 2024-05 | | MGSM (multilingual math) | 91.5% | [OpenAI GPT-4o system card](https://openai.com/index/gpt-4o-system-card/) | 2024-05 | | MathVista | 63.8% | [OpenAI GPT-4o system card](https://openai.com/index/gpt-4o-system-card/) | 2024-05 | *Scores are reported by OpenAI and shown for context. They are not Benchgen measurements. See the system card for full harness details.* ## GPT-4o vs Alternatives | Model | Context | GPQA Diamond | HumanEval | Price (in/out per 1M) | |---|---|---|---|---| | **GPT-4o** | 128K | 53.6% | 90.2% | $2.50 / $10 | | [Claude 3.5 Sonnet](/models/anthropic/claude-3-5-sonnet) | 200K | 65.0% | 93.7% | $3 / $15 | | [Gemini 1.5 Pro](/models/google/gemini-1-5-pro) | 2M | 46.2% | — | $1.25 / $5 | | [GPT-5](/models/openai/gpt-5) | 400K | — | — | $1.25 / $10 | Claude 3.5 Sonnet edges GPT-4o on GPQA and HumanEval at a modestly higher price; Gemini 1.5 Pro offers a dramatically larger context window at lower cost; GPT-5 supersedes GPT-4o on most tasks at the same output price. GPT-4o's main advantages remain ecosystem breadth, audio nativety, and the free ChatGPT tier for prototyping. *(Rival scores from respective provider announcements; not Benchgen measurements.)* ## How GPT-4o Performs on Real Agent Tasks GPT-4o's agent performance is well-characterised in the industry: strong instruction-following, reliable tool call formatting, and consistent JSON structured output (especially after the November 2024 update). The 128K context is adequate for most multi-step agent workflows, and the native vision capability makes it a natural fit for agents that need to read screenshots, parse PDFs with visual structure, or interact with GUIs. Where GPT-4o shows gaps is in very long-horizon planning tasks — workflows requiring 20+ decision steps, deep multi-document reasoning, or chained tool calls with complex state — where the o-series reasoning models (o1, o3) or GPT-5 show meaningful advantages. For Benchgen evaluations, teams running GPT-4o as an agent backbone typically see solid Tool-Call Accuracy and Skill Coverage scores, with more variance on Goal Completion and Regression Stability for long, multi-session workflows. ## Use GPT-4o via API ```python from openai import OpenAI client = OpenAI(api_key="YOUR_API_KEY") response = client.chat.completions.create( model="gpt-4o-2024-11-20", messages=[{"role": "user", "content": "Analyse this contract clause for risk..."}], max_tokens=1024, ) print(response.choices[0].message.content) ``` ## Related Models & Benchmarks - Compare: [Claude 3.5 Sonnet](/models/anthropic/claude-3-5-sonnet) · [Claude Opus 4.6](/models/anthropic/claude-opus-4-6) · [GPT-5](/models/openai/gpt-5) · [Gemini 1.5 Pro](/models/google/gemini-1-5-pro) - Benchmarks: [SWE-bench Verified](/benchmarks/swe-bench-verified) · [MMLU-Pro](/benchmarks/mmlu-pro) ## Frequently Asked Questions

What is GPT-4o?

GPT-4o is OpenAI's omni flagship model, released May 13, 2024. It processes text, vision, and audio natively in a single neural network — the first GPT model to handle all three modalities end-to-end without a pipeline of separate models.

What is GPT-4o's context window?

128,000 tokens — roughly 200 pages of text in a single request — with a maximum output of 16,384 tokens per response.

How much does GPT-4o cost?

$2.50 per million input tokens and $10 per million output tokens. Cached input is $1.25/1M; batch processing halves both rates. Free access is available in ChatGPT (with rate limits).

Is GPT-4o open source?

No. GPT-4o is a proprietary, closed model available only via OpenAI's API, ChatGPT, and Azure OpenAI. Weights are not released.

What is GPT-4o's knowledge cutoff?

October 2023. Data and events after that date are not in the model's base knowledge and require retrieval augmentation.

What is the difference between GPT-4o and GPT-4o mini?

GPT-4o mini is a smaller, cheaper distillation ($0.15/$0.60 per 1M) optimised for high-volume, lower-complexity tasks. GPT-4o is the full flagship with stronger reasoning, better vision, and audio capabilities.

--- *Specs and scores sourced from [OpenAI's GPT-4o announcement](https://openai.com/index/hello-gpt-4o/) and [GPT-4o system card](https://openai.com/index/gpt-4o-system-card/). Third-party benchmark scores attributed inline. Last updated 2026-06-19.* # GPT-4o GPT-4o is a large language model developed by OpenAI.

GPT-4o — Evaluations

Evaluations