As many of you may have already noticed, GPT-5.4 and GPT-5.4 Pro are now available inside AI Builder.
As usual, we try to make new models available to Pickaxe builders as soon as they are released, so you can start experimenting right away without waiting for long platform updates.
If you’ve been building for a while, the biggest difference you’ll likely notice with 5.4 is how consistently it follows instructions. Prompts with multiple steps, formatting rules, or structured outputs tend to behave more reliably compared to earlier versions.
A few places where builders might see improvements:
• Pickaxes with long or complex prompts
• Tools that require step-by-step reasoning
• Bots that must stick to a format or tone
• Knowledge-heavy tools pulling from uploaded documents
GPT-5.4 Pro is the higher precision variant. It’s designed for more demanding tasks like deeper analysis, research workflows, or tools that require more careful reasoning.
Where to find it in Pickaxe
Both models are already available in the AI Model selector inside the Builder.
Steps:
Open your Pickaxe in the Builder
Go to the Prompt tab
Scroll to AI Model
Select GPT-5.4 or GPT-5.4 Pro
If anyone has already tested it with their tools, would love to hear what differences you’re seeing. Some Pickaxes improve noticeably just by switching models.
One big change in the new models is the context windows are now 1 million tokens which is a big improvement and in line with Gemini and Claude (Grok is still bigger at 2 million tokens for the Grok 4.1 Fast Reasoning model and smaller for Grok 4).
For anyone watching costs, here’s a quick token pricing increase with this model.
Model
Input
Output
Change vs Previous
GPT-4o (common old)
$2.50
$10.00
—
GPT-5.2 (recent)
$1.75
$14.00
—
GPT-5.4 (new)
$2.50
$15.00
+43% input vs 5.2 +50% output vs 4o
GPT-5.4 Pro (new)
$30.00
$180.00
Premium tier (was ~$21/$168 on prior Pro)
The new standard model is a bit more expensive overall (especially on output tokens), and the Pro version is a clear step up in price for heavy reasoning/research tasks. That said, the stronger consistency and reasoning often means fewer total tokens or retries, which can help offset the difference in practice.
Has anyone run the same prompt side-by-side yet and seen how the actual credit usage compares? Would love to hear real numbers!
In my own small tests comparing for example Gemini 3 Flash and GPT 5.4 the price difference is somewhere 10-30x (depending on how much “research” need to be done) meaning if Gemini 3 Flash costs $0.01 then GPT 5.4 is costing around $0.10-0.30. Only difference a lot of time seems to be that Gemini understands to keep the answer shorter while GPT 5.4 likes to go really in detail on some topics where simpler answer would been enough.
A lot of time users don’t want to even read mile long answer so keeping things shorter and more condensed is usually better UX (at least in my opinion).
