OpenAI Puts GPT-5.4 Mini in Front of Every Free User
OpenAI extended GPT-5.4 Mini to free-tier ChatGPT users on March 17, accessible via the Thinking toggle, while simultaneously retiring GPT-5 Thinking Mini within 30 days. The move consolidates a sprawling model lineup and pushes capable reasoning to the widest audience yet.
OpenAI moved the reasoning baseline on March 17. GPT-5.4 Mini is now available to every free-tier ChatGPT user through the Thinking toggle in the model picker — no subscription required.
The model significantly outperforms GPT-5 Mini on coding and reasoning benchmarks, running more than 2x faster while approaching GPT-5.4’s performance on SWE-Bench Pro and OSWorld-Verified. That is meaningful. SWE-Bench Pro measures real software engineering task completion, not toy problems. Pushing a model of this caliber to the free tier is not a minor quality-of-life update.
At the same time, OpenAI announced GPT-5 Thinking Mini will be deprecated within 30 days. The company is also rolling out a simplified model picker across Plus, Pro, Business, and Enterprise tiers, collapsing a menu that had grown complicated enough to confuse users who just wanted to pick the best available model.
The full GPT-5.4 — available to paid tiers — includes a 1-million-token context window and support for autonomous multi-step workflow execution across software environments. That positions it directly against Claude’s computer use capabilities and Google Gemini’s long-context workloads.
Free users do still hit rate limits. After 10 messages with GPT-5.3 per five-hour window, the system falls back to a mini model. But the ceiling has moved up — the mini model free users get as a fallback is now meaningfully better than anything that existed at the free tier a year ago.
GPT-5.4 Mini is also live in the API and in Codex, at lower per-token pricing than the full model. GPT-5.4 Nano, a smaller variant, is available solely through the API.
OpenAI is reportedly approaching $25 billion in annualized revenue and exploring a late-2026 IPO. The product cadence supports that story: the company is not slowing down on model releases and is actively expanding who can access frontier-class reasoning. The competitive pressure from Anthropic, Google, and now Xiaomi is showing up in the pace of free-tier improvements. Users benefit from that fight regardless of who they think is winning it.