OpenAI has released two new models — GPT-5.4 mini and GPT-5.4 nano — designed for speed, cost efficiency, and multi-agent workflows. GPT-5.4 mini is now live for free ChatGPT users, marking one of the most significant expansions of flagship-tier AI to the general public so far.
What’s New
Both models are compact versions of the full GPT-5.4, released two weeks ago. They’re not stripped-down fallbacks — they’re purpose-built for workloads where latency matters more than raw reasoning depth.
- GPT-5.4 mini: More than 2x faster than GPT-5 mini, with major improvements across coding, reasoning, multimodal understanding, and tool use. Approaches GPT-5.4 performance on SWE-Bench Pro and OSWorld-Verified benchmarks. Available in ChatGPT Free, Go, and Pro tiers, plus the API and Codex.
- GPT-5.4 nano: The smallest and cheapest model in the lineup, priced at roughly $0.20 per million input tokens. API-only for now. Built for classification, data extraction, ranking, and lightweight subagent tasks.
Why This Release Is Different
The framing around these models signals a shift in how OpenAI thinks about AI architecture. Rather than a single model handling everything, GPT-5.4 mini and nano are explicitly designed to work inside multi-agent systems — where a larger model plans and orchestrates, and smaller subagents execute specific tasks in parallel at speed.
In OpenAI’s own Codex platform, for example, GPT-5.4 handles planning and final judgment while GPT-5.4 mini subagents handle codebase search, large file review, and document processing simultaneously. This pattern is becoming the standard architecture for production AI systems.
What It Means for Developers
For developers building on the API, the cost difference is significant. The full GPT-5.4 costs $2.50 per million input tokens. GPT-5.4 nano comes in at $0.20 — a 12x reduction. For high-volume pipelines where speed and cost matter, that gap changes what’s economically viable to build.
GPT-5.4 mini is also now rolling out in GitHub Copilot across VS Code, JetBrains, Visual Studio, and Xcode — available to Pro, Pro+, Business, and Enterprise users via the model picker.
Performance on Coding Tasks
On GPQA Diamond (a test of expert-level reasoning), GPT-5.4 mini scored 88.01% — trailing the full GPT-5.4’s 93% by a relatively small margin. On coding benchmarks, it approaches flagship-level pass rates while running significantly faster. Notion’s engineering team reported that it often matches or beats more expensive models on complex formatting tasks at a fraction of the compute cost.
Conclusion
GPT-5.4 mini and nano aren’t just cheaper versions of GPT-5.4 — they’re the foundation of a new way of building AI systems, where speed and cost efficiency at the subagent level matter as much as raw intelligence at the top. If you’re building with AI or just using ChatGPT on the free tier, the experience just got meaningfully faster. Browse our AI coding tools directory to see how GPT-5.4 fits into the wider ecosystem.
