AI Potluck
Model components / Base / pretrained models

GPT-4o

OpenAI

OpenAI's 2024 multimodal foundation model, natively handling text, image and audio with a 128K context and an October 2023 knowledge cutoff. It is API-only with no released weights, and has been superseded by the GPT-5 line; it is being retired as the ChatGPT default in favor of GPT-5.5 Instant.

OpenAI's model page carries no benchmarks, so the capability figures are launch-era external numbers. The adoption band rests on a ChatGPT weekly-actives figure re-sourced 2026-08-14 after the original aggregator began blocking automated reads. Verified 2026-08-13 via OpenAI's GPT-4o model documentation.

Openness

1 high confidence
1.0
weights
closed
data
closed
code
closed
license
Proprietary(API-only)

No downloadable weights: GPT-4o is available only through the OpenAI API and ChatGPT, the same position as the rest of OpenAI's model line.

  • https://developers.openai.com/api/docs/models/gpt-4o recorded 2026-08-13

    API-only model page - "our most capable model outside of our o-series models", 128,000 context window, Oct 01, 2023 knowledge cutoff, text and image input. No weight download, no corpus, no training code.

Adoption

5 medium confidence
5.0

GPT-4o was the default ChatGPT model for much of 2024 and 2025, before GPT-5. ChatGPT crossed roughly 900 million weekly active users in February 2026 and about 1 billion in May 2026, relayed from OpenAI's own announcement rather than from an aggregator estimate. The level is attributed honestly to the ChatGPT surface GPT-4o powered rather than to a standalone GPT-4o count, and on either reading it sits two orders of magnitude above the >10M threshold.

Capability

3 medium confidence
3.0

Frontier in 2024, but surpassed by the 2026 reasoning frontier - GPT-5 and Gemini 3.5 both score 5 - and a model from mid-2025 or earlier that has been surpassed scores 3 rather than 5. One caveat on the evidence: OpenAI's current documentation page carries no benchmark figures at all, only the description of GPT-4o as 'our most capable model outside of our o-series models', so what is sourced here is the positioning. The launch-era MMLU of about 88.7 and MMMU of about 69.1 come from external write-ups and have no primary benchmark source behind them.

Verified 2026-08-13