GPT-4

Claims and evidence

• Released 14 March 2023Released in ChatGPT Plus and to developers from an API waitlist, with text input only at first. ● Primary source[1]ChatGPT Plus; API [4]March 14, 2023: Announcing GPT-4 in ChatGPT
• Deprecated 22 April 2026 ● Primary source[5]2026-04-22: Legacy GPT model snapshots
• gpt-4-0613 with function calling released 13 June 2023 ● Primary source[6]GPT-4
• GPT-4 API generally available to existing paying developers 6 July 2023 ● Primary source[5]2023-07-06: GPT and embeddings [7]GPT-4 API general availability
• Image input (GPT-4 with vision) rolling out in ChatGPT 25 September 2023The system card describes GPT-4 with vision as a capability being made broadly available; the ChatGPT release notes of the same day announce image input for Plus users. ● Primary source[3]Abstract [4]September 25, 2023: New voice and image capabilities in ChatGPT
• Retired from ChatGPT and replaced there by GPT-4o 30 April 2025Date announced in advance in the release note of April 10, 2025. ● Primary source[4]April 10, 2025: Sunsetting GPT-4 in ChatGPT

• Status Deprecatedgpt-4 (snapshot gpt-4-0613) is scheduled to shut down in the API on October 23, 2026. gpt-4-0314 was shut down on 2026-03-26 and the gpt-4-32k snapshots on 2025-06-06. GPT-4 was retired from ChatGPT on 2025-04-30. ● Primary source[5]2026-04-22: Legacy GPT model snapshots [4]April 10, 2025: Sunsetting GPT-4 in ChatGPT
• Replaced by GPT-5.6 SolSubstitute model listed for gpt-4-0613 on the deprecations page when checked on 2026-10-01. ● Primary source[5]2026-04-22: Legacy GPT model snapshots

• Successor of GPT-3.5The research post compares GPT-4 with GPT-3.5 throughout and describes GPT-3.5 as the first test run of the training stack later used for GPT-4. ● Primary source[1]first paragraphs; Capabilities [2]Abstract; Section 4

• Change · Modality Accepts image input in addition to text; at launch image input was tested with a single partner, and it reached ChatGPT in September 2023.Compared with GPT-3.5 ● Primary source[1] [3]
• Change · Safety OpenAI reports that after six months of alignment work GPT-4 responds less often to requests for disallowed content than GPT-3.5.Compared with GPT-3.5 ● Primary source[1]
• Change · Other OpenAI says GPT-4 is more reliable and creative than GPT-3.5 and handles more nuanced instructions once tasks become complex.Compared with GPT-3.5 ● Primary source[1]

• Input text, imageText input was released first; image input started in limited alpha and reached ChatGPT as GPT-4 with vision from 2023-09-25. The API model gpt-4 accepts text only according to its model page. ● Primary source[1]first paragraph; Visual inputs [2]Abstract
• Output text ● Primary source[1]first paragraph [8]Model details
• Feature Function callingFrom gpt-4-0613 (2023-06-13). ● Primary source[6]Function calling; GPT-4
• Context window 8,192 tokensThe gpt-4-32k variant has 32,768 tokens. ● Primary source[1]API [8]Model details
• Parameters Not disclosedThe technical report says it contains no further details about the architecture, including model size. ● Primary source[2]Section 2
• Access consumer app, API ● Primary source[1]ChatGPT Plus; API

Lineage

Variants

Related AI Radar coverage

AI Radar coverage starts in June 2026; no coverage linked yet.

All model releases from OpenAI on AI Radar →