Article
March 5, 2026
Introducing GPT-5.3 Instant, GPT-5.4 Thinking, and GPT-5.4 Pro
# Work Users
The latest upgrade to the GPT-5 series
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.
- GPT-5.3 Instant for fast, everyday work
- GPT-5.4 Thinking for difficult, real-world professional tasks and longer workflows
- 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.
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.
Can you help me write a response to a customer who is upset about [issue]?Help me write a response to a customer who is upset about [issue]. Keep it professional and compliant. Don’t add a preamble—start with the draft. If you can’t do something directly, offer the closest safe alternative.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.
Search the web for the latest on [topic]. Then give me:Bottom line (2–3 sentences)What changed recently vs what’s unchangedWhy it matters for a [team/function] team3 risks or misconceptions to watch forThe 3 most credible sources you used- Ask the same prompt but change only one variable: “for a finance team” vs “for a support team.”
- 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.
Rewrite this for a work audience. Goals: clearer, shorter, and more direct. Keep the meaning the same.
Output:Version 1: conciseVersion 2: slightly warmer but still professionalThen list 5 specific edits you made (not general advice)
[paste text]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.
Draft an internal announcement about [change].
Requirements: practical, not marketing-y, and easy to skim with short paragraphs.
Give me two drafts:Draft A: straightforward and conciseDraft B: more narrative, but still professional
Then provide a one-paragraph “what I’d say live” talk track.
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.
I’m trying to complete [task].Context: [paste relevant details].Requirements: [constraints, stakeholders, tone, format].Do this in one pass:Ask up to 3 clarifying questions (only if truly necessary)Propose a planProduce the first draft outputList what you’d verify if you had access to my sources/tools
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.
Help me run this workflow end-to-end: [workflow].
List:the tools/systems you’d use at each stepwhat you would look for to verify successlikely failure points and how you’d detect them
Then draft the final deliverable: [doc/email/plan].
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.
Create a decision memo on [topic].
Audience: [exec/team].
Include:Recommendation (top)Options considered + tradeoffsRisks + mitigationsWhat would change your mindDraft comms plan (internal + external)
Availability
- GPT-5.3 Instant is available in ChatGPT for Plus, Pro, Business, and Enterprise/Edu (and rolls out to additional tiers shortly after).
- GPT-5.4 Thinking is available in ChatGPT (Plus, Pro, Business, Enterprise) and is also available to developers in the API and Codex.
- 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)
- GPT-5.3 in the API: gpt-5.3-chat-latest
- 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:
- Early access governs availability of newly released models during rollout.
- Legacy models governs continued access to older models (when available) for a limited period.