October 23 is the scheduled shutdown date for several older API models, including GPT-4 Turbo, the original GPT-4o snapshot, and selected o1 and o3-mini versions. Developers using affected model IDs should migrate to supported replacements to avoid service disruption.
Designed for classification, routing, and other tasks with bounded answers, it supports text, images, structured inputs, and up to 400 questions per request. OpenAI reports up to 10x faster responses compared with GPT-6 Luna through the Responses API.
The new SDK combines speech recognition, generation, reasoning, and tool calling for interactive voice applications. Transcription supports more than 50 languages, with the models currently available in public preview.
Developers can now keep shared prompts cached for at least 12 hours, reducing costs when reusing context across requests. Cache writes cost 2.5 times the normal input price, so savings depend on how often the cached content is reused.
According to MiniMax, the new model delivers nearly 3x faster inference and 70% lower task costs than its predecessor. It targets coding, computer use, and multi-step agent workflows, and is available through the API and coding tools.