GPT-5

Claims and evidence

• Released 7 August 2025 ● Primary source[1]page date; Availability and access [2]page date; Availability & pricing [4]Aug 7, 2025
• Deprecated 11 June 2026 ● Primary source[5]2026-06-11: GPT-5 and o3 model deprecations
• GPT-5 pro released in the API 6 October 2025 ● Primary source[4]Oct 6, 2025

• Status Deprecated ● Primary source[5]2026-06-11: GPT-5 and o3 model deprecations
• Replaced by GPT-5.6 SolRecommended replacement for gpt-5-2025-08-07 on the deprecations page when read on 2026-10-01 (gpt-5.6-terra for the mini snapshot, gpt-5.6-luna for nano). ● Primary source[5]2026-06-11: GPT-5 and o3 model deprecations

• Successor of GPT-4oThe system card lists gpt-5-main as the successor of GPT-4o; GPT-5 replaced GPT-4o as the ChatGPT default. ● Primary source[6]Table 1: Model progressions [1]How to use GPT-5
• Successor of o3The system card lists gpt-5-thinking as the successor of o3; the o3 docs say o3 is succeeded by GPT-5. ● Primary source[6]Table 1: Model progressions [7]introduction
• Successor of o4-miniThrough the inline variant GPT-5 mini: the system card lists gpt-5-thinking-mini as the successor of o4-mini, and the o4-mini docs say it is succeeded by GPT-5 mini. ● Primary source[6]Table 1: Model progressions [8]introduction
• Successor of GPT-4.1The system card lists gpt-5-thinking-nano (GPT-5 nano) as the successor of GPT-4.1 nano, an inline variant of GPT-4.1; GPT-5 also replaced GPT-4.1 in ChatGPT. ● Primary source[6]Table 1: Model progressions [1]How to use GPT-5
• Successor of o3-proThrough the inline variant GPT-5 pro, which the announcement says replaces o3-pro; the system card maps gpt-5-thinking-pro to o3 Pro. ● Primary source[6]Table 1: Model progressions [1]GPT-5 pro

• Change · Reasoning In ChatGPT a router chooses between a fast model and a reasoning model, instead of separate model choices.Compared with GPT-4o ● Primary source[1]
• Change · Efficiency OpenAI says GPT-5 with thinking needs 50 to 80 percent fewer output tokens than o3 for better results across several capabilities.Compared with o3 ● Primary source[1]
• Change · Safety Safe-completions training aims for the most helpful answer within safety limits, instead of a plain choice between complying and refusing.Compared with o3 ● Primary source[6]
• Change · Other OpenAI reports fewer factual errors and less deception than o3 in its own evaluations.Compared with o3 ● Primary source[1]

• Input text, image ● Primary source[3]Model details
• Output text ● Primary source[3]Model details
• Context window 400,000 tokensThe docs give 128,000 maximum output tokens; the developer post gives a maximum of 272,000 input tokens and 128,000 reasoning and output tokens. ● Primary source[3]Model details [2]Availability & pricing
• Parameters Not disclosedNot disclosed by OpenAI. No claim madeNo source
• Access consumer app, API ● Primary source[1]Availability and access [2]Availability & pricing [4]Aug 7, 2025

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 →