What are GPT-5.4 mini and nano models?

GPT-5.4 mini and nano are the latest additions to OpenAI's GPT-5 model family, launched on March seventeenth and eighteenth, twenty twenty-six respectively. GPT-5.4 mini is designed for production workloads requiring fast inference and cost efficiency, running more than two times faster than the previous GPT-5 mini while maintaining comparable quality. GPT-5.4 nano is an even smaller model optimized specifically for classification, data extraction, ranking, and coding subagent tasks where full model intelligence is unnecessary. Both models expand OpenAI's offering across the performance-cost spectrum.

How much does GPT-5.4 mini cost to use?

GPT-5.4 mini is priced at zero point seven five dollars per million input tokens and four point five zero dollars per million output tokens through the OpenAI API. This represents a significant cost reduction compared to the full GPT-5.4 model and makes advanced AI capabilities more accessible for production applications. For context, processing one million tokens of input is roughly equivalent to reading several hundred pages of text. Developers building applications that require frequent API calls can see substantial cost savings by routing appropriate queries to the mini model rather than using the more expensive full-size models.

Can free ChatGPT users access GPT-5.4 models?

Yes, free ChatGPT users now have access to GPT-5.4 mini through the new Thinking feature in ChatGPT. This is a significant expansion of free tier capabilities, as previous thinking and reasoning features were limited to paid ChatGPT Plus and Pro subscribers. The Thinking mode allows free users to have the model reason through complex problems step by step before providing a final answer, which improves accuracy on math, logic, and analytical questions. However, free users have usage limits on the Thinking feature and cannot access the full GPT-5.4 model or nano through the chat interface.

How does GPT-5.4 mini compare to GPT-5 mini?

GPT-5.4 mini offers substantial improvements over the previous GPT-5 mini in both speed and efficiency. OpenAI reports that GPT-5.4 mini runs more than two times faster than GPT-5 mini while maintaining comparable quality across most benchmarks. The newer model also uses only thirty percent of the GPT-5.4 quota allocation when used in Codex, making it more resource-efficient for coding tasks. OpenAI has announced that GPT-5 Thinking mini will be retired within thirty days of the GPT-5.4 mini launch, suggesting they view the newer model as a direct and superior replacement.

What is GPT-5.4 nano designed for?

GPT-5.4 nano is specifically designed for lightweight AI tasks that do not require the full reasoning capabilities of larger models. Its primary use cases include text classification where documents or messages need to be categorized into predefined groups, data extraction where structured information needs to be pulled from unstructured text, ranking tasks where items need to be ordered by relevance or priority, and serving as coding subagents within larger AI systems. The nano model trades some general capability for extreme speed and cost efficiency, making it suitable for high-volume production pipelines where latency and cost are critical constraints.

GPT-5.4 Mini & Nano 2026: Free Thinking Mode for All Users

OpenAI 2026: $840B Valuation, IPO Plans & GPT-5.4 Launch

openai@release:~/models$ cat gpt-5.4-release.md

OpenAI launched GPT-5.4 mini at $0.75/1M tokens — 2x faster — March 17-18, 2026. Free ChatGPT users now access Thinking mode. Here is what changed.

Published: March 23, 2026

GPT-5.4 miniNEW

MMLU: 88.7%
Speed: 2x faster than GPT-5 mini
Price: $0.75 / $3.00 per 1M tokens

GPT-5.4 nanoNEW

MMLU: 79.2%
Optimized for: on-device & edge
Price: $0.15 / $0.60 per 1M tokens

Free UsersUPGRADE

Thinking mode: unlocked
Previously: Plus-only
Democratizing AI access

OpenAI launches GPT-5.4 mini and nano models, expands Thinking mode to free users

Photo: 9to5Mac — ChatGPT app icon

0x

Mini Speed

faster than GPT-5 mini

0.0%

MMLU Score

GPT-5.4 mini benchmark

$0.00

Nano Input

per 1M input tokens

0M+

Free Users

get Thinking mode

Model Comparison

OpenAI launched GPT-5.4 mini on March 17, 2026, followed by GPT-5.4 nano on March 18. Both models prioritize cost efficiency, aiming to expand AI access for developers and small businesses. Below is a detailed comparison based on data from OpenAI Blog, March 17, 2026.

Model	Speed	MMLU	Input/1M	Output/1M	Context	Best For
GPT-5.4 miniNEW	2x faster than GPT-5 mini	88.7%	$0.75	$3.00	128K	Chatbots, summarization, analysis
GPT-5.4 nanoNEW	Smallest, fastest model	79.2%	$0.15	$0.60	32K	Edge computing, on-device
GPT-5 mini	Previous generation	85.1%	$1.50	$6.00	128K	Similar to 5.4 mini
GPT-4o mini	Legacy generation	82.0%	$0.30	$1.20	128K	Being phased out

API Cost Calculator

Input tokens: 1.0M

Output tokens: 0.5M

Input cost$0.7500
Output cost$1.5000
Total$2.2500

MMLU Benchmark Scores

Source: OpenAI Blog — March 17, 2026

GPT-5.4 mini88.7%

GPT-5 mini85.1%

GPT-4o mini82%

GPT-5.4 nano79.2%

Claude 3.5 Haiku77.8%

Gemini 2.0 Flash76.5%

GPT-5.4 mini scored 88.7% on the MMLU benchmark, surpassing GPT-5 mini (85.1%) despite costing half as much. GPT-5.4 nano achieved 79.2%, a remarkable result for a model optimized for on-device and edge computing. According to The Verge's assessment (March 18, 2026), this represents a significant step in the AI pricing race.

GPT-5.4 mini performance comparison with previous models

Photo: The Verge — GPT-5.4 mini benchmark performance

What Changes for Free Users

Alongside the new model launches, OpenAI unlocked Thinking mode for all free ChatGPT users. Previously, this feature was exclusive to Plus subscribers ($20/month). This is part of OpenAI's broader strategy to democratize AI access, according to the OpenAI Blog announcement on March 17, 2026.

// Features unlocked for Free tier

Thinking modePlus onlyAll users

Message limit10 msgs/3hrs25 msgs/3hrs

Default modelGPT-4o miniGPT-5.4 mini

Timeline

March 17, 2026

GPT-5.4 mini launched

OpenAI announced GPT-5.4 mini on its official blog, highlighting that the model is 2x faster than GPT-5 mini while cutting costs by 50%. MMLU score reached 88.7%, the highest in the small model segment.

→ If your startup spends $3,000/month on GPT-5 mini API, switching to 5.4 mini could save ~$1,500/month with better performance.

March 18, 2026

GPT-5.4 nano launched

One day after mini, OpenAI followed with GPT-5.4 nano — the company's smallest and cheapest model. At just $0.15/1M input tokens, nano is optimized for edge computing and mobile devices.

→ At $0.15/1M tokens, a chatbot app processing 10 million messages/day would cost roughly $1.5/day for input tokens alone.

March 18, 2026

Thinking mode opens for all

OpenAI unlocked Thinking mode (advanced reasoning) for free users. Previously exclusive to Plus subscribers ($20/month). The message limit was also raised from 10 to 25 messages per 3 hours.

→ Over 100 million free users now get advanced reasoning — equivalent to saving $20/month if you previously subscribed to Plus just for Thinking mode.

Developer Impact

The launch of GPT-5.4 mini and nano marks a significant shift in the AI development ecosystem. With dramatically lower costs, startups and independent developers can build AI products without massive budgets. Per ZestLab analysis, these areas stand to benefit the most:

SaaS Applications

50-80% lower API costs enable small SaaS products to integrate AI without overspending. Nano is particularly suited for classification and quick summarization tasks.

Edge & IoT

GPT-5.4 nano with 32K context and small footprint is suited for running directly on end devices, from smartphones to IoT hardware.

Education & Research

Nano pricing ($0.15/1M) enables schools and labs to deploy AI at scale with limited budgets.

Free ChatGPT users get Thinking mode access

Photo: TechCrunch — Thinking mode unlocked for free users

Impact on Vietnam's AI Ecosystem

Vietnam has over 2,000 tech startups, many of which use OpenAI APIs. The launch of GPT-5.4 mini and nano at significantly lower prices will make AI more accessible for Vietnamese businesses. Per ZestLab analysis, the biggest beneficiaries in Vietnam include: e-commerce (customer support chatbots), education (AI tutoring), and fintech (automated credit scoring).

GPT-5.4 nano is particularly suited for the Vietnamese market — where many users access the internet via mobile phones. Nano's on-device capability opens opportunities for offline-first applications, serving users in rural areas.

OpenAI — All News and Analysis OpenAI Valuation 2026 — Detailed Analysis GPT-5.4 Launch — Overview

References

OpenAI Blog — Introducing GPT-5.4 mini and nano — 17/3/2026
The Verge — GPT-5.4 benchmark and performance review — 18/3/2026
TechCrunch — OpenAI unlocks Thinking mode for free users — 18/3/2026

Benchmark scores and pricing per OpenAI official announcement on March 17, 2026. Impact analysis by ZestLab.

By Hoa Dinh · Founder & Senior Tech Editor

Published: March 23, 2026 · Updated: March 25, 2026

Explore ZestLab Tools

JSON Formatter Base64 Encoder Markdown Preview

technology·GPT-5.4 mini · GPT-5.4 nano · OpenAI March 2026 · ChatGPT free thinking