GPT-4o
A model by OpenAI · released 2024-05-13 · 75 checked facts from 16 sources
Timeline
- OpenAI released GPT-4o, May 13, 2024
- GPT-4o is available on ChatGPT
- GPT-4o is available on API
- OpenAI deprecates GPT-4o
Prices
Per price, each value with the dates it was seen. A value seen on one date only is one copy of the page.
1 price changed over the record.
| Price for | Values over time |
|---|---|
| per 1M cached input tokens (audio, realtime) | $2.50 |
| per 1M cached input tokens (fine-tuning) | $1.875 |
| per 1M cached input tokens (standard) | $1.25 |
| per 1M cached input tokens (text, realtime) | $2.50 |
| per 1M input tokens (audio, realtime) | $40 |
| per 1M input tokens (batch) | $1.25 |
| per 1M input tokens (fine-tuning) | $3.75 |
| per 1M input tokens (standard) | $5 → $2.50 |
| per 1M input tokens (text, realtime) | $5 |
| per 1M output tokens (audio, realtime) | $80 |
| per 1M output tokens (batch) | $5 |
| per 1M output tokens (fine-tuning) | $15 |
| per 1M output tokens (standard) | $10 |
| per 1M output tokens (text, realtime) | $20 |
| per 1M tokens (fine-tuning training) | $25 |
| per 1M tokens (vision, fixed) | $5 |
| image (low resolution) | $0.001275 |
Dated versions
Pinned snapshots of the model, priced on their own.
gpt-4o-2024-05-13
| per 1M input tokens (standard) | $5 |
| per 1M output tokens (batch) | $7.50 |
| per 1M output tokens (standard) | $15 |
gpt-4o-2024-08-06
| per 1M cached input tokens (fine-tuning standard) | $1.875 |
| per 1M input tokens (batch) | $1.875 |
| per 1M input tokens (fine-tuning batch) | $1.875 |
| per 1M input tokens (fine-tuning standard) | $3.75 |
| per 1M output tokens (batch) | $7.50 |
| per 1M output tokens (fine-tuning batch) | $7.50 |
| per 1M output tokens (fine-tuning standard) | $15 |
| per 1M tokens (fine-tuning training) | $25 |
| per 1M tokens (training) | $25 |
Scores
Self-reported: stated on the lab's own channel. A score reported twice with two values is shown twice, with
its date: a lab re-runs a benchmark under a new setup; neither replaces the other.
| Benchmark | Score | Date | Who |
|---|---|---|---|
| AIME '24 | 13.1% | 2025-04-14 | self-reported |
| AIME ‘24 (math) | 9.3% | 2025-02-28 | self-reported |
| AIME’24 | 9.3% | 2025-02-19 | reported by xAI |
| ARC-AGI | 50% | 2026-10-01 | reported by stoppels.ch |
| Aider’s polyglot | 18.2% | 2025-04-14 | self-reported |
| CharXiv-D | 85.3% | 2025-04-14 | self-reported |
| CharXiv-Reasoning | 52.7% | 2025-04-14 | self-reported |
| EgoSchema | 72.2% | 2025-02-19 | reported by xAI |
| GPQA (science) | 53.6% | 2025-02-19 | self-reported |
| GPQA Diamond | 46.0% | 2025-04-14 | self-reported |
| Graphwalks | 41.7% | 2025-04-14 | self-reported |
| IFEval | 81.0% | 2025-04-14 | self-reported |
| International Mathematics Olympiad (IMO) | 13 | 2024-09-12 | self-reported |
| LOFT (128k) | 78.0% | 2025-02-19 | reported by xAI |
| LiveCodeBench | 32.3% | 2025-02-19 | reported by xAI |
| MMLU | 87.2% | 2024-05-13 | self-reported |
| MMLU | 85.7% | 2025-04-14 | self-reported |
| MMLU-pro | 72.6% | 2025-02-19 | reported by xAI |
| MMMLU (multilingual) | 81.5% | 2025-02-28 | self-reported |
| MMMU | 69.1% | 2025-02-19 | reported by xAI |
| MMMU | 68.7% | 2025-04-14 | self-reported |
| MMMU (multimodal) | 69.1% | 2025-02-28 | self-reported |
| MathVista | 61.4% | 2025-04-14 | self-reported |
| MultiChallenge | 27.8% | 2025-04-14 | self-reported |
| Multilingual MMLU | 81.4% | 2025-04-14 | self-reported |
| SWE-Bench Verified (coding) | 30.7% | 2025-02-28 | self-reported |
| SWE-Lancer Diamond (coding) | 23.3% | 2025-02-28 | self-reported |
| SWE-bench Verified | 33.2% | 2025-04-14 | self-reported |
| Script concordance testing (SCT) | 63.9% | 2026-10-26 | reported by PubMed: AI in top medical journals |
| SimpleQA | 38.2% | 2025-02-19 | reported by xAI |
| Video-MME | 65.3% | 2025-04-14 | self-reported |
| jailbreaking test | 22 | 2024-09-12 | self-reported |
What it does
- GPT-4o: can reason across audio, vision, and text in real time
- GPT-4o: accepts as input any combination of text, audio, and image and generates any combination of text, audio, and image outputs
- GPT-4o: matches GPT-4 Turbo performance on text in English and code
- GPT-4o: significant improvement on text in non-English languages
- GPT-4o: audio response latency (minimum) 232 milliseconds
- GPT-4o: audio response latency (average) 320 milliseconds
- GPT-4o: price down 50% vs GPT-4 Turbo
- GPT-4o: speed up 2x vs GPT-4 Turbo
- GPT-4o: rate limits up 5x vs GPT-4 Turbo
- GPT-4o: trained a single new model end-to-end across text, vision, and audio
- GPT-4o: better at vision and audio understanding compared to existing models
- GPT-4o: sets new high watermarks on multilingual, audio, and vision capabilities
- GPT-4o: external red teaming experts 70+
- GPT-4o: 128K context window
- GPT-4o: vision capabilities
- GPT-4o: multimodal
- GPT-4o: faster and cheaper than GPT-4 Turbo with stronger vision capabilities
- GPT-4o: SWE-Lancer Diamond earnings $138,750
- GPT-4o: 128,000 context window
- GPT-4o: share of GPT-generated factual accuracy grades within the full range of human scores 94%
In the news
- The Atlantic: The Quest to Build an Immortal AI Companion
- cnet.com: Man Sues OpenAI After Wife Allegedly Used ChatGPT to Help Poison Him
- digitalhealthwire.com: OpenAI in the Crosshairs of Wrongful Death Lawsuit
- gizmodo.com: Seattle Woman Died After Following ChatGPT's Medical Advice
- lawcommentary.com: OpenAI Sued After ChatGPT Allegedly Told Woman to Stop Seizure Drug Before She Died
- oxygen.com: Family Alleges ChatGPT Told Woman to Stop Taking Epilepsy Medication Before Her Death
- bankinfosecurity.com: Lawsuit: ChatGPT Dispensed Fatal Medical Advice to Woman
- cnet.com: Woman Died After ChatGPT Advised Her to Stop Taking Medication, Lawsuit Alleges
- kiro7.com: Seattle woman’s family sues OpenAI, claims ChatGPT coached her off medication before death
- sea.mashable.com: Lawsuit says ChatGPT persuaded person to stop taking epilepsy meds, leading to her death
- americanbazaaronline.com: OpenAI sued by parents who say ChatGPT coached daughter to stop seizure medication before her death
- finance.yahoo.com: Social Media Victims Law Center and The Schenk Law Firm File Wrongful Death Lawsuit Against OpenAI Alleging ChatGPT’s Medical Advice Resulted in the Death of a Woman with Epilepsy
Built 2026-11-21 06:50 UTC from the news
knowledge graph by code. Every line is a fact whose quote was checked against the source's own words (hover a source);
it says who states it, not that it is true. Archived pages are the lab's pages as the Wayback Machine holds them.