Model Details
GPT-4o (May 2024)
Quick answer: GPT-4o (May 2024) is OpenAI's original "omni" model, released May 13, 2024. The first model to natively process and generate text, vision, and audio in a single end-to-end model, it replaced GPT-4 Turbo as the default API model and dropped prices 50% from its predecessor. At $2.50/$10 per million tokens (after OpenAI's mid-2024 price reduction), it remains a capable and cost-effective choice for production workloads not requiring frontier reasoning.
At a Glance
Where GPT-4o (May 2024) leads
- First truly native omni model: text, vision, and audio processed in one network — no separate vision adapter
- 2× price reduction vs GPT-4 Turbo at launch, later reduced again to $2.50/$10
- Faster response times than GPT-4 Turbo with comparable quality on most tasks
- Strong vision understanding: charts, screenshots, diagrams, and document OCR
Where it lags
- 128K context window (GPT-5 and Claude Opus 4.6 offer 400K–1M)
- SWE-bench Verified ~33% at launch — substantially below the 70%+ frontier of 2025–2026
- Superseded by GPT-4o (Nov 2024), GPT-4.5, and GPT-5 for new builds
Best for: production workloads on pinned model versions, vision-heavy pipelines, and teams with existing GPT-4o integrations not yet ready to migrate.
What GPT-4o (May 2024) Is
GPT-4o — the "o" stands for "omni" — was OpenAI's breakthrough in unified multimodality. Previous OpenAI models handled vision through a separate adapter bolted onto the text model. GPT-4o was trained end-to-end across modalities, enabling lower latency, richer cross-modal reasoning, and native audio understanding.
At launch it also introduced a real-time audio API that allowed voice conversations with human-like response latency (232ms average), a sharp contrast to the 2–3 second round-trip of previous voice integrations. The May 2024 checkpoint is identified by the API string gpt-4o-2024-05-13 and is still accessible for teams that pinned this version.
Specifications
| Field | Value |
|---|---|
| Organization | OpenAI |
| API identifier | gpt-4o-2024-05-13 |
| Context window | 128,000 tokens |
| Max output | 4,096 tokens |
| License | Proprietary |
| Release date | May 13, 2024 |
| Knowledge cutoff | October 2023 |
| Modality | Multimodal (text, vision, and audio input; text output) |
Pricing
| Input (per 1M tokens) | Output (per 1M tokens) | |
|---|---|---|
| OpenAI | $2.50 | $10.00 |
Price reduced from the original $5/$15 launch pricing. Source: OpenAI pricing.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| SWE-bench Verified | 33.2% | OpenAI — Hello GPT-4o | 2024-05 |
| MMLU | 88.7% | OpenAI — Hello GPT-4o | 2024-05 |
| GPQA | 53.6% | OpenAI — Hello GPT-4o | 2024-05 |
| HumanEval | 90.2% | OpenAI — Hello GPT-4o | 2024-05 |
Scores reported by OpenAI. Not Benchgen measurements.
GPT-4o vs Alternatives
| Model | Context | SWE-bench Verified | Price (in/out per 1M) |
|---|---|---|---|
| GPT-4o (May 2024) | 128K | 33.2% | $2.50 / $10 |
| GPT-4o (Nov 2024) | 128K | 38.8% | $2.50 / $10 |
| GPT-5 | 400K | 74.9% | $1.25 / $10 |
| Claude 3.5 Sonnet | 200K | 49.0% | $3 / $15 |
(Rival scores from respective announcements; not Benchgen measurements.)
Related Models & Benchmarks
- Compare: GPT-4o (Nov 2024) · GPT-5 · Claude 3.5 Sonnet
- Benchmarks: SWE-bench Verified · MMLU-Pro
Frequently Asked Questions
What is GPT-4o?
GPT-4o ("omni") is OpenAI's first natively multimodal model, processing text, vision, and audio in a single network. The May 2024 checkpoint was the original release, available via API asgpt-4o-2024-05-13. How much does GPT-4o cost?
$2.50 per million input tokens and $10 per million output tokens (after mid-2024 price reduction from the original $5/$15 launch pricing).Specs sourced from OpenAI's GPT-4o launch. Last updated 2026-06-19.