On April 3, 2026, OpenAI retired GPT-4o from all plans. On April 23, GPT-5.5 launched — described by OpenAI as its “smartest and most intuitive model yet,” with particular strength in agentic coding, computer use, and knowledge work. Four days later, OpenAI shipped a redesigned model picker that puts thinking effort controls directly in the message composer.
In three weeks, OpenAI’s entire front-facing model stack changed.
If you are still mentally referencing GPT-4o or the older reasoning models, you are using an outdated map. This guide gives you the current one: every model that exists today, what each is actually for, when the quality difference between tiers genuinely matters, and the decision framework that picks the right model in under 30 seconds.
🔗 This is Post #2 in the ChatGPT Unlocked series. Start with ChatGPT Masterclass 2026 if you are new. Later posts cover Advanced Voice Mode, Memory, Images 2.0, and Custom GPTs in depth.
The Current ChatGPT Model Family (April 2026)
GPT-5.3 Instant
The fast everyday tier
GPT-5.3 Instant is the speed-optimized model for high-frequency, lower-complexity tasks. It handles most conversational queries, common factual questions, straightforward writing tasks, and everyday lookups with minimal latency.
It also powers Fast answers — ChatGPT’s newest UX feature that identifies common factual queries and returns rapid, high-confidence responses without routing through a full generation cycle. Ask “What is the capital of France?” and Fast answers intercepts it before GPT-5.3 even fully initializes.
Plan availability: Free, Go, Plus, Pro, Business, Enterprise (often the fallback when other models hit rate limits)
When to use it:
- Quick factual lookups
- Conversational back-and-forth
- Simple editing tasks (fix the grammar in this paragraph)
- High-volume batch tasks via the API where cost per call matters
- When you need an answer in under 5 seconds and the question is not complex
When NOT to use it:
- Anything requiring multi-step reasoning
- Complex creative work requiring sustained tone and voice
- Analysis of long or technical documents
- Tasks where small reasoning errors cascade
GPT-5.4 Thinking
The professional workhorse
Released in March 2026, GPT-5.4 Thinking is the model most Plus users will use for the majority of serious work. OpenAI described its goals as delivering “complex real work accurately, effectively, and efficiently” — and the framing is accurate. GPT-5.4 Thinking incorporates capabilities from GPT-5.3 Codex (the coding-specialized variant) while improving document creation, spreadsheet reasoning, and multi-tool workflows.
The word “Thinking” in the name reflects the chain-of-thought reasoning built into its generation. Unlike GPT-5.3, which responds quickly, GPT-5.4 Thinking pauses to work through problems internally before responding. The pause is often only a few seconds but produces qualitatively different results on tasks requiring judgment.
Plan availability: Plus, Pro, Business, Enterprise
When to use it:
- Professional writing (reports, proposals, client communications)
- Coding projects with real complexity
- Data analysis and interpretation
- Long-form content requiring sustained quality
- Research synthesis across multiple sources
- Most business tasks that are not trivially simple
When NOT to use it:
- Quick factual lookups where GPT-5.3 is faster
- Tasks that genuinely require GPT-5.5’s frontier capabilities (you are paying message allocation for marginal gain)
GPT-5.4 Pro
Maximum GPT-5.4 capability
GPT-5.4 Pro is the ceiling of the GPT-5.4 generation — highest accuracy, deepest reasoning, most reliable outputs. The practical difference between GPT-5.4 Thinking and GPT-5.4 Pro is most visible on tasks that sit at the edge of GPT-5.4 Thinking’s reliable range: very long documents, highly technical domains, complex multi-step coding architecture.
Plan availability: Pro, Business, Enterprise only
When to use it:
- When GPT-5.4 Thinking produces inconsistent results on a specific task
- Very long technical documents
- High-stakes outputs where maximum accuracy matters and you are on a Pro plan
- As a cost-efficient alternative to GPT-5.5 Pro for many tasks
GPT-5.5 (April 23, 2026 — current frontier)
The new standard for capable AI
GPT-5.5 is OpenAI’s most significant model release of 2026. The key quality is not simply more intelligence — it is a fundamentally more agentic character. GPT-5.5 understands what you are trying to accomplish at a goal level, not just at an instruction level. It plans, uses tools, checks its own work, navigates unexpected intermediate states, and continues working until the task is complete.
OpenAI President Greg Brockman: “GPT-5.5 is a real step forward towards the kind of computing that we expect in the future.”
The domains where this matters most: agentic coding (writing and debugging multi-file projects), computer use (navigating interfaces to complete tasks), knowledge work (research → synthesis → polished document in one session), and early scientific reasoning.
Plan availability: Plus, Pro, Business, Enterprise
When to use it:
- Multi-step projects that previously required many back-and-forth iterations
- Complex coding tasks involving multiple files or systems
- Research projects requiring sustained focus across many sources
- Tasks where you want to describe a goal and have ChatGPT execute rather than step-by-step instruct
- Anything that felt frustratingly close with GPT-5.4 but never quite landed
When NOT to use it:
- Simple everyday tasks (wasted on GPT-5.5)
- High-volume API tasks where cost per call matters
- When GPT-5.4 Thinking already handles your task reliably
GPT-5.5 Thinking
Extended reasoning at the frontier
GPT-5.5 Thinking adds extended chain-of-thought to the GPT-5.5 base — the model reasons through a problem step-by-step internally, producing a visible thinking trace before the final response. The thinking budget is adjustable via the new thinking effort controls in the model picker.
This is the model for genuinely hard problems: complex mathematical reasoning, nuanced multi-variable analysis, contested research questions, and architectural decisions where small errors have large downstream consequences.
Plan availability: Plus, Pro, Business, Enterprise (with thinking effort controls)
The thinking effort levels (April 28, 2026 addition):
- Standard: Lightweight reasoning pass. Faster, lower message cost.
- Thinking: Balanced reasoning. Recommended default for most uses of this model tier.
- Extended: Maximum reasoning depth. Noticeably slower, use for genuinely hard problems.
GPT-5.5 Pro
Unrestricted frontier performance
GPT-5.5 Pro is the maximum-capability ChatGPT model. Rate limits are significantly higher, model behavior is optimized for the most demanding tasks, and access is not throttled during peak usage. This is what OpenAI’s Pro plan ($200/month) primarily unlocks.
Plan availability: Pro, Business, Enterprise only
When it makes a difference: High-stakes research where GPT-5.5 still makes reasoning errors, extended agentic tasks running for many minutes, and professional contexts where the marginal quality gain justifies the additional plan cost.
GPT-5.4 mini
Cost-efficient reasoning fallback
GPT-5.4 mini is a smaller, faster, cheaper variant of the GPT-5.4 generation. In the ChatGPT interface, it serves as the rate-limit fallback when you exhaust your allocation on higher models. In the API, it is a cost-efficient option for tasks that benefit from reasoning capability but need to run at high volume.
Plan availability: Free, Go (as the Thinking option), and as fallback/API option across all paid plans.
The Thinking Effort Controls: A New Dial
As of April 28, 2026, when you select GPT-5.5 Thinking, GPT-5.4 Thinking, or any Thinking-tier model in the model picker, three effort controls appear:
Standard → Quick reasoning pass. Use for: tasks where you want the reasoning architecture but not the full depth. Roughly equivalent to asking a smart person to think briefly before answering.
Thinking → Balanced reasoning. The default. Use for: most analytical tasks. Visibly better than Standard on complex problems without the latency of Extended.
Extended → Maximum reasoning depth. Use for: genuinely hard problems where you have tried Standard and Thinking and still need more. The visible thinking trace is longest here; the response takes longer but the quality ceiling is higher.
Practical rule: Start at Thinking for analytical work. Move to Extended only when you can clearly articulate why the task requires deeper reasoning than Thinking provides. Using Extended for everything wastes time and message allocation.
The 30-Second Decision Framework
Use this before selecting a model:
Is this a simple factual question or quick lookup?
→ Let Fast answers handle it, or GPT-5.3 Instant
Is this a professional writing, analysis,
or coding task of moderate complexity?
→ GPT-5.4 Thinking (Thinking effort level)
Is this complex enough that GPT-5.4 produces
inconsistent or incomplete results?
→ GPT-5.5
Does this require extended reasoning — math,
research synthesis, architectural decisions,
genuinely contested analysis?
→ GPT-5.5 Thinking (Extended effort if needed)
Are you on Pro plan and need maximum accuracy
on the hardest possible task?
→ GPT-5.5 Pro
The framework reads from cheapest-and-fastest to most-capable. Start at the top and move down only when the task genuinely requires it.
Task-by-Task Model Recommendations
| Task | Best Model | Why |
|---|---|---|
| Quick factual question | Fast answers / GPT-5.3 | Speed, no reasoning needed |
| Grammar and editing | GPT-5.3 Instant | Overkill for GPT-5.5 |
| Email drafting | GPT-5.4 Thinking | Balance of quality and speed |
| Professional proposal | GPT-5.4 Thinking or GPT-5.5 | Sustained quality required |
| Complex code debugging | GPT-5.5 | Multi-step reasoning, tool use |
| Data analysis from upload | GPT-5.5 | Extended code execution |
| Market research synthesis | GPT-5.5 Thinking | Cross-source reasoning |
| Mathematical reasoning | GPT-5.5 Thinking (Extended) | Highest accuracy ceiling |
| Strategic decision analysis | GPT-5.5 Thinking | Multi-variable reasoning |
| Creative fiction | GPT-5.4 or GPT-5.5 | Depends on complexity |
| Image generation | Any model + Images 2.0 | Model routes to image pipeline |
| Agentic multi-step project | GPT-5.5 | Core GPT-5.5 strength |
Model Access by Plan: Quick Reference
| Model | Free | Go | Plus | Pro | Business/Enterprise |
|---|---|---|---|---|---|
| GPT-5.3 Instant | ✅ | ✅ | ✅ | ✅ | ✅ |
| Fast answers | ✅ | ✅ | ✅ | ✅ | ✅ |
| GPT-5.4 mini | Limited | ✅ | Fallback | Fallback | ✅ |
| GPT-5.4 Thinking | ❌ | ❌ | ✅ | ✅ | ✅ |
| GPT-5.4 Pro | ❌ | ❌ | ❌ | ✅ | ✅ |
| GPT-5.5 | ❌ | ❌ | ✅ | ✅ | ✅ |
| GPT-5.5 Thinking | ❌ | ❌ | ✅ | ✅ | ✅ |
| GPT-5.5 Pro | ❌ | ❌ | ❌ | ✅ | ✅ |
API Model Strings (For Developers)
If you are building with the OpenAI API, use these model identifiers. Exact strings should be verified at platform.openai.com/docs/models — OpenAI updates these with each release.
# Current model strings (verify at platform.openai.com)
"gpt-5.3" # GPT-5.3 Instant
"gpt-5.4" # GPT-5.4 Thinking
"gpt-5.4-pro" # GPT-5.4 Pro
"gpt-5.5" # GPT-5.5
"gpt-5.5-thinking" # GPT-5.5 Thinking
"gpt-5.5-pro" # GPT-5.5 Pro
"gpt-5.4-mini" # GPT-5.4 mini
API pricing note: Reasoning models with extended thinking consume additional tokens for the thinking trace. In the API, you can configure reasoning_effort as "low", "medium", or "high" — mirroring the Standard / Thinking / Extended controls in the ChatGPT interface.
What Changed From the Old Model Family
For users coming from the GPT-4o era:
| Old (retired) | Current equivalent | What changed |
|---|---|---|
| GPT-4o | GPT-5.4 Thinking | Significantly stronger, reasoning-native |
| GPT-4o mini | GPT-5.3 Instant / GPT-5.4 mini | Faster, cheaper |
| o1 | GPT-5.5 Thinking (Thinking effort) | Integrated, no separate model selector |
| o1-pro | GPT-5.5 Thinking (Extended) | Now accessible via effort controls |
| o3 | GPT-5.5 Pro | Folded into unified tier |
The old model picker that required selecting between “GPT-4o,” “o1,” and “o3” as separate named models is gone. The new unified system selects capability level and reasoning depth through the effort controls rather than requiring you to know which internal architecture maps to your task.
Common Model Selection Mistakes
Mistake 1: Using GPT-5.5 for everything on Plus GPT-5.5 is rate-limited on Plus. Using it for tasks that GPT-5.4 Thinking handles equally well burns your allocation faster. Save GPT-5.5 for tasks that genuinely need it.
Mistake 2: Never using Extended thinking Extended thinking on hard analytical problems produces meaningfully better results. Users who avoid it because it is slower are leaving quality on the table for the tasks where it matters most.
Mistake 3: Using Standard effort on genuinely complex reasoning The Standard effort level is not appropriate for complex multi-step problems. If you are analyzing research, working through a strategic decision, or debugging a subtle logical error — use Thinking or Extended.
Mistake 4: Not recognizing that Fast answers have their own ceiling Fast answers are optimized for speed and confidence on common queries. For nuanced questions that look factual but involve contested evidence or recent developments, use web search + a full model rather than relying on Fast answers.
Conclusion
The ChatGPT model family has changed more in April 2026 than in any comparable period. GPT-5.5’s arrival, the retirement of GPT-4o, and the redesigned model picker with thinking effort controls together represent a step-change in what the platform offers — and how you access it.
The framework is now simpler than the old named-model system: pick the right capability tier for the task, set the reasoning depth for that tier, and let the model handle the rest. The model picker’s effort controls are not a complication — they are a new degree of control that was not previously available.
Your next step: Open ChatGPT’s model picker. Find GPT-5.5. Select Thinking effort. Give it a task that has previously felt like it was almost working but not quite. See what the current frontier looks like.
📚 Continue the Series:
- ← Previous ChatGPT Masterclass 2026
- Next → ChatGPT Advanced Voice Mode: Talking to AI Like It’s a Person
- For building The OpenAI API for Non-Developers
- For deep reasoning OpenAI Reasoning Models: When AI Thinks Before It Answers
Last updated: May 2026. OpenAI model releases are frequent — verify the current model lineup at platform.openai.com/docs/models and the ChatGPT model picker before making architecture decisions.
