OpenAI Launches GPT-5.4 Mini & Nano: Faster, Smarter AI—Even for Free Users

We are used to massive AI flagship models recording the headlines for their deep “thinking” capabilities. However, the real-world utility of AI often depends on something very simple: speed. This week, OpenAI addressed this directly by launching GPT-5.4 mini and GPT-5.4 nano, two compact models designed to bring near-flagship performance to everyday tasks without the wait.

The most exciting news for the general public is that GPT-5.4 mini is now available to Free and Go tier users of ChatGPT. According to OpenAI, users can access a model twice as fast as GPT-5 mini via the “Thinking” option. Plus, it offers significantly improved reasoning and coding skills.

Meet OpenAI’s GPT-5.4 Mini and GPT-5.4 Nano

What makes these “mini” and “nano” versions stand out isn’t just that they are small—it’s how closely they mimic their larger siblings. In several key benchmarks, GPT-5.4 mini performed remarkably close to the full-scale GPT-5.4. For instance, on the GPQA Diamond test (which measures high-level expertise), the mini model scored 88.01%, trailing the flagship’s 93% by a surprisingly small margin.

These models also excel in “latency-sensitive” environments. This includes coding assistants that need to feel instantaneous or systems that interpret screenshots and dense user interfaces in real time.

The rise of the “subagent”

An interesting shift highlighted in this release is the concept of AI subagents. Think of it like a professional hierarchy: a large, highly intelligent model (like GPT-5.4 Thinking) acts as a senior manager, handling the complex planning and final judgment. It then delegates narrower, repetitive tasks—like searching a codebase or reviewing a massive document—to the faster, cheaper GPT-5.4 mini or nano.

External tech platforms are already embracing this strategy. Notion, a well-known productivity app, has discovered that the GPT-5.4 mini can handle focused tasks with great accuracy. Abhisek Modi, Notion’s AI engineering lead, said that the model often matches or beats more expensive versions when it comes to handling complex formatting, all while using a fraction of the computing power.

Availability

OpenAI is making these tools widely accessible through several channels. To start, ChatGPT users will find it in the Free and Go tiers via the “Thinking” feature. For paid users, it serves as a high-speed fallback when they reach their flagship limits.

For developers, both models are live in the OpenAI API. They will find a staggering cost difference: while the full GPT-5.4 costs $2.50 per million input tokens, the nano version is priced at just $0.20. Lastly, coders can enjoy the capabilities of GPT-5.4 mini via the Codex app and IDE extensions. This allows developers to handle simpler debugging and navigation loops at about one-third of the usual “quota” cost.