Overview
GPT-5.4 mini and GPT-5.4 nano are OpenAI's newest small models, released on March 17, 2026. They bring the capabilities of GPT-5.4 down to faster and cheaper variants designed for high-volume, latency-sensitive, and multi-agent workflows. GPT-5.4 mini is OpenAI's primary recommendation for coding assistants, subagents, and interactive applications that need performance close to the full GPT-5.4 without the associated cost. GPT-5.4 nano is the lightest and cheapest model in the GPT-5.4 family, targeted at classification, data extraction, and ranking tasks where speed and cost matter more than deep reasoning.
Together, they mark a significant step beyond GPT-5 mini: both benchmarks and API pricing reflect a new tier of small-model capability that makes it practical to run GPT-5.4-class intelligence at scale.
What's New
GPT-5.4 mini: Stronger Coding, Faster Response
GPT-5.4 mini is a major step up from GPT-5 mini across coding, reasoning, multimodal understanding, and tool use. It achieves 54.4% on SWE-Bench Pro (up from 45.7% for GPT-5 mini), approaching the 57.7% of the full GPT-5.4, and hits 72.1% on OSWorld-Verified compared to 42.0% for GPT-5 mini—a 30-point improvement in computer use via screenshot interpretation. Despite the quality gains, it runs more than 2x faster than its predecessor.
GPT-5.4 mini is now available to ChatGPT Free and Go users via the "Thinking" feature in the + menu, making it the most capable model OpenAI has ever made available at no cost.
GPT-5.4 nano: Cheapest Model for Subagents
GPT-5.4 nano is OpenAI's smallest, cheapest GPT-5.4 variant. It is API-only and costs $0.20 per 1M input tokens and $1.25 per 1M output tokens—the lowest price in the GPT-5.4 family. OpenAI positions it explicitly for subagent roles: classification, data extraction, ranking, and other simpler supporting tasks that do not require the full reasoning depth of mini or the flagship model.
Codex can now delegate less reasoning-intensive tasks to GPT-5.4 mini subagents internally, and developers can mirror this pattern in their own pipelines by routing lightweight tasks to nano and complex tasks to mini or full GPT-5.4.
400K Context Window (mini)
GPT-5.4 mini has a 400K-token context window, matching GPT-5 mini. GPT-5.4 nano also has a 400K-token context window. This enables longer document processing, extended coding sessions, and richer multi-turn agent context without switching to the full GPT-5.4 model.
GitHub Copilot Integration
GPT-5.4 mini is now generally available in GitHub Copilot and is rolling out across supported Copilot surfaces.
Performance Benchmarks
| Benchmark | GPT-5 mini | GPT-5.4 mini | GPT-5.4 (full) |
|---|---|---|---|
| SWE-Bench Pro | 45.7% | 54.4% | 57.7% |
| OSWorld-Verified | 42.0% | 72.1% | — |
| Speed vs GPT-5 mini | baseline | 2x+ faster | — |
GPT-5.4 mini's OSWorld-Verified score of 72.1%—a 30-point jump over GPT-5 mini—is particularly notable for applications that rely on computer use and screenshot-based task execution.
Compatibility Notes
GPT-5.4 mini availability:
- ChatGPT Free and Go: available via the "Thinking" feature in the + menu. For Plus and Pro users, GPT-5.4 mini currently serves as a rate-limit fallback for GPT-5.4 Thinking rather than a separately listed model option.
- OpenAI API: available (
gpt-5.4-mini) - GitHub Copilot: generally available as of March 17, 2026
- Codex: available; Codex can delegate subagent tasks to mini internally
GPT-5.4 nano availability:
- OpenAI API only (
gpt-5.4-nano) - Not available in the ChatGPT consumer interface
- Not available in GitHub Copilot at launch
GPT-5.4 nano's API-only access means it is a developer-facing model. Teams looking for the cheapest path to GPT-5.4-class capabilities in a chatbot or productivity UI should use GPT-5.4 mini instead.
Pricing & Plans
ChatGPT Plans (GPT-5.4 mini)
| Plan | Price | GPT-5.4 mini Access |
|---|---|---|
| Free | $0/month | ✅ via Thinking feature |
| Go | $8/month (US; pricing localized in some markets) | ✅ via Thinking feature |
| Plus | $20/month | ✅ |
| Pro | $200/month | ✅ + extended limits |
API Pricing
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| GPT-5.4 mini | $0.75 | $4.50 |
| GPT-5.4 nano | $0.20 | $1.25 |
GPT-5.4 mini is 3x more expensive per input token than GPT-5 mini ($0.25), and GPT-5.4 nano is approximately 4x pricier than its predecessor on input. The price increase reflects the substantial benchmark improvements over the previous generation.
Best For
- Developers building AI agent pipelines that need cost-efficient, high-quality subagents for coding and reasoning tasks
- ChatGPT Free and Go users who want access to GPT-5.4-class reasoning without a paid subscription
- Enterprise teams running high-volume classification, extraction, or ranking pipelines via the API, where GPT-5.4 nano's pricing makes scale economical
- GitHub Copilot users who want the latest OpenAI coding model available directly in their IDE
- AI code generator workflows where GPT-5.4 mini's SWE-Bench Pro score closes the gap with full flagship models
- Product teams integrating multi-model architectures that route complex tasks to full GPT-5.4 and delegate lighter subtasks to nano
FAQ
Is GPT-5.4 mini available to ChatGPT Free users?
Yes. GPT-5.4 mini is available to ChatGPT Free and Go users via the "Thinking" feature in the model selector. This makes it the most capable model OpenAI has ever offered at no cost. Full GPT-5.4 remains restricted to paid tiers.
What is GPT-5.4 nano, and how is it different from mini?
GPT-5.4 nano is smaller, faster, and cheaper than mini, designed for subagent roles like classification, data extraction, and ranking. It is only available via the OpenAI API—not in the ChatGPT interface or GitHub Copilot. GPT-5.4 mini is the better choice for more general-purpose tasks, interactive applications, and scenarios requiring stronger reasoning.
How do the new models compare to the previous ChatGPT 5.2 release?
GPT-5.4 mini and nano follow the full GPT-5.4 release trajectory, extending the 5.4 architecture to smaller variants. GPT-5.4 mini specifically improves coding and computer-use scores dramatically over GPT-5 mini, the equivalent small model from the GPT-5 generation.
Are these models suitable for use in Codex?
Yes. GPT-5.4 mini is available in Codex, and OpenAI says Codex can delegate simpler supporting tasks to GPT-5.4 mini subagents to reduce cost while keeping complex reasoning on the full model. GPT-5.4 nano is API-only and is not described as a Codex model at launch.
Will API pricing increase for existing GPT-5 mini users?
GPT-5.4 mini is priced at $0.75/1M input tokens, versus $0.25/1M for GPT-5 mini—a 3x increase. Teams with high-volume workloads on GPT-5 mini should evaluate whether the benchmark improvements justify the cost increase before migrating. GPT-5 mini remains available in the API for teams that prioritize cost over the performance uplift.
