We’ve recently released three new models in the in the GPT-5 family—so you can pick the right tradeoff between speed and depth.

1. GPT-5.3 Instant for fast, everyday work
2. GPT-5.4 Thinking for difficult, real-world professional tasks and longer workflows
3. GPT-5.4 Pro for the most demanding work where you want the highest-capability option available

GPT-5.3 Instant: more helpful, more natural everyday conversations

It’s tuned to make everyday chats feel easier: more accurate answers, better web results, and fewer moments where the conversation stalls out or gets weighed down by caveats.

View our blog post.

GPT-5.3 Instant is tuned to get to a safe, useful answer more directly. In practice, this helps when you’re doing real work—drafting customer responses, writing internal comms, summarizing policies, or asking for guidance on a sensitive topic where a normal, compliant answer is appropriate. Use the same question twice: once “plain,” once with the guardrails. Compare how quickly you get to something usable.

GPT-5.3 Instant also improves how it searches and uses information from the web. It’s better at balancing what it finds online with its own reasoning—for example, using existing understanding to contextualize recent updates rather than simply summarizing results.

More broadly, it’s less likely to return a long list of links. It does a stronger job of recognizing what you’re really asking and surfacing the most important information upfront, resulting in answers that are more immediately usable without sacrificing speed or tone. Pick a topic where recency matters (a competitor change, a regulation update, a product announcement), and ask for a decision-ready summary.

1. Ask the same prompt but change only one variable: “for a finance team” vs “for a support team.”
2. You should see the same facts, but a different emphasis and “so what.”

A smoother, more to-the-point conversational style

This update has a more focused yet natural conversational style. It cuts back on unnecessary proclamations and stays closer to the task you’re trying to complete.

As always, you can adjust the way the model responds—like its warmth and enthusiasm—within settings. Use ChatGPT as a “tight editor” for something you actually send.

Stronger creative writing, with more range and texture

GPT-5.3 Instant is also better at writing that needs voice and variation—useful for internal narratives, launch notes, scripts, and long-form drafts where you want the writing to feel less templated while staying clear. Turn a dry outline into two distinct drafts you can choose from.

GPT-5.4 Thinking: for difficult, real-world professional tasks

GPT-5.4 Thinking is designed for work that’s harder to do in a single pass: multi-step reasoning, long context, tool-heavy workflows, and outputs that need to be accurate and usable. In ChatGPT, it’s the smartest and most efficient model yet for difficult professional tasks—better at delivering what you actually asked for with less back and forth.

It also brings stronger coding capabilities into the flagship frontier model, and is built to handle longer, tool-heavy workflows with lower latency.

Better at long, tool-heavy workflows

When your task spans multiple steps—gather context, use tools, verify, then produce a final output—GPT-5.4 Thinking is tuned to stay coherent and complete the workflow more reliably. Use it for a work task that normally takes multiple “back-and-forth” turns.

Improved computer use and tool search (for agentic tasks)

GPT-5.4 introduces improved computer-use capabilities, enabling agents to interact with software to complete and verify real tasks. It also improves “tool search,” helping agents scale to larger tool ecosystems without paying a tax in tokens or intelligence. Give it a workflow that involves tools/systems, and ask for a verifiable plan plus checks.

GPT-5.4 Pro: premium option for the most demanding work

GPT-5.4 Pro is the highest-cost, highest-capability option in this set. Use it for tasks where you’re optimizing for quality and depth over speed—especially when the output needs to be decision-ready or hard to revise after the fact. Use it when the output needs to be “exec-safe” with low iteration.

Availability

1. GPT-5.3 Instant is available in ChatGPT for Plus, Pro, Business, and Enterprise/Edu (and rolls out to additional tiers shortly after).
2. GPT-5.4 Thinking is available in ChatGPT (Plus, Pro, Business, Enterprise) and is also available to developers in the API and Codex.
3. GPT-5.4 Pro is available to Pro and Enterprise users, and is also available to developers in the API and Codex.

API model IDs (for developers)

1. GPT-5.3 in the API: gpt-5.3-chat-latest
2. GPT-5.4 in the API: gpt-5.4 and gpt-5.4-pro

Admin FAQ

What is the “Early access” toggle?

Early access is an admin control that determines whether users in your workspace can access newly released models during the early access window. This is now a persistent toggle.

What happens when Early access is ON?

Once enabled, eligible users can automatically receive new models as they launch and see them in the model picker during the early access window.

What happens when Early access is OFF?

Users won’t see early access models in their model picker until the early access period ends (typically about 7 days post-launch).

Where do I manage Early access?

In workspace admin settings under Permissions + Roles.

How does Early access relate to Legacy Models?

They’re separate:

1. Early access governs availability of newly released models during rollout.
2. Legacy models governs continued access to older models (when available) for a limited period.