GPT-5.4 mini reaches free ChatGPT users: what changes

OpenAI has begun rolling out GPT-5.4 mini to free and Go plan users inside ChatGPT, bringing reasoning capabilities to the broadest tier of the platform for the first time. The model is accessible through the Thinking option in the plus (+) menu and represents a significant step down in the barrier to entry for AI reasoning tools.

For paid subscribers, the update works differently. GPT-5.4 mini now serves as an automatic fallback when users on Plus, Pro, or other paid tiers hit GPT-5.4 Thinking rate limits. Rather than facing a hard stop or reverting to a non-reasoning model, the system keeps reasoning available during high-traffic periods. Enterprise customers retain the option to route Auto mode to GPT-5.4 mini by default if that fits their workflow better. The model will not appear as a selectable option in the model picker, and GPT-5 Thinking mini will be retired as a standalone choice within 30 days.

The rollout arrives shortly after OpenAI confirmed that became the platform’s flagship professional model on March 5, introducing a 1 million token context window and native computer use capabilities. Making a lighter version of that reasoning tier available to free users continues a pattern OpenAI has followed with previous model generations, where scaled-down variants eventually reach the widest possible audience.

Alongside the mini rollout, OpenAI also announced the removal of the legacy Deep Research mode, scheduled for March 26. The change affects only the older version of the feature. The current Deep Research experience stays intact, and all historical conversations and results remain accessible. Users who relied on the legacy mode for their research workflows will need to transition to the updated version ahead of the deadline.

The broader context here is competitive pressure. [INTERNAL LINK: AI chatbots comparison] Claude Sonnet 4.6 became the free default on Claude.ai in February, and Google has been expanding Gemini access across its free tiers throughout the quarter. Making reasoning available without a paywall is increasingly a baseline expectation rather than a premium differentiator.

You may be interested in

Siemens Eigen Engineering Agent interface for industrial automation workflows

Siemens Eigen Engineering Agent: AI for industrial automation

AI agent automates industrial engineering design and validation workflows.
April 21, 2026

Google TurboQuant: the ICLR 2026 efficiency breakthrough explained

New compression method cuts memory overhead. Big implications for on-device AI.
April 12, 2026

GPT-6 release date: what OpenAI confirmed in April 2026

Pretraining confirmed complete. Release expected weeks away, says Altman.
April 11, 2026
We recommend
AI focused on thoughtful analysis, coding support, and research tasks
AI code assistant with agent mode. Best for professional developers.
8.8
Conversational AI for learning through supportive, empathetic dialogues
8.8
AI creating original songs from simple text prompts
AI calendar assistant defending focus time, optimizing schedules