OpenAI released GPT-5.4 mini and nano on March 17, bringing the capabilities of their latest flagship model to smaller, faster packages. The catch: these “budget” models now cost up to three times more than their predecessors.
What’s New
GPT-5.4 mini runs more than twice as fast as GPT-5 mini while approaching the full GPT-5.4’s performance on key benchmarks. It scores 54.4% on SWE-Bench Pro (compared to 57.7% for the full model) and 72.1% on OSWorld-Verified (vs. 75%). That’s a massive jump from GPT-5 mini’s 45.7% and 42.0% respectively.
GPT-5.4 nano is designed for high-volume, simple tasks: classification, data extraction, ranking, and basic coding. It’s the fastest model in OpenAI’s lineup, hitting around 200 tokens per second.
Both models support:
- 400,000-token context windows
- Text and image inputs
- Tool use and function calling
- Web search and file search
- Computer use capabilities
The Price Jump
Here’s where it gets painful for developers:
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| GPT-5.4 mini | $0.75 | $4.50 |
| GPT-5 mini (old) | $0.25 | $1.50 |
| GPT-5.4 nano | $0.20 | $1.25 |
GPT-5.4 mini costs 3x more than its predecessor. The Hacker News thread captured developer sentiment: “the lowest end pricing has been thoroughly hiked.”
For comparison, Gemini 3 Flash costs $0.50/$3.00 and Claude Haiku 4.5 sits at $0.80/$4.00. OpenAI’s mini is no longer the obvious budget choice.
Where They Shine (And Don’t)
Developers on Hacker News reported mixed results. Speed improvements are real—responses that took 15 seconds on GPT-5 mini now complete in 2-3 seconds. But some noticed consistency issues: “5.4 mini seems to struggle with consistency, and even with temperature 0 sometimes gives the correct response, sometimes a wrong one.”
For agentic workflows, several developers noted GPT models underperform Claude. One comment described GPT as “robotic, mechanic, constantly mis-using tools” compared to Claude’s reliability for autonomous agents.
Nano shines for high-volume classification tasks. One developer mentioned using it for image classification “millions of times” where cost per call matters more than peak capability.
Availability
- GPT-5.4 mini: Available now in ChatGPT (Free and Plus users get it as a “Thinking” option), Codex, and the API
- GPT-5.4 nano: API only
In Codex, mini uses about 30% of the standard GPT-5.4 quota, making it roughly a third of the cost for integrated development work.
What This Means
OpenAI is betting developers will pay more for faster, more capable small models. The gamble makes sense—mini’s benchmark scores approach the full GPT-5.4, which itself costs $10/$30 per million tokens.
But the price hike comes at an awkward time. Gemini Flash and open-weight models like Llama keep getting better while staying cheap or free. For basic tasks, the value proposition of OpenAI’s smallest models just got harder to justify.
If you’re building high-volume applications, do the math carefully. That 3x price increase adds up fast at scale.