All articles
IT & Technology

ChatGPT's Model Turnover in 2026: What Changed and What to Use Now

OpenAI shipped seven GPT-5 variants in under a year. Here is what each model does, what got retired, and which one you should actually pick.

  • #chatgpt
  • #gpt-5
  • #openai
  • #ai-models
  • #model-comparison
chatgpt-model-turnover-2026

If you last paid attention to ChatGPT in early 2025, the model you benchmarked is probably gone. GPT-4o is retired. The o3 reasoning model is retired. Even the original GPT-5.0 from August 2025 has dropped off OpenAI’s API pricing page [S3, S4].

In under a year, OpenAI shipped seven distinct versions of GPT-5, each with different capabilities, context windows, and price points [3]. The pace has been roughly one new model or major feature every six weeks [6]. For anyone using ChatGPT casually or building on the API, the practical question is no longer whether to upgrade but which model to pick right now.

How the GPT-5 Family Evolved

Each GPT-5 variant solved a specific problem, and understanding that progression helps explain what you are actually getting today.

GPT-5.0 launched on August 7, 2025 as a unified system with a fast base model and a deeper reasoning layer that activated automatically based on query complexity [S3, S4]. It scored 94.6% on AIME 2025 math without tools and 100% with Python tools in Pro mode [3].

GPT-5.1 arrived on November 13, 2025 with adaptive reasoning that made it 2 to 3x faster on simple tasks while keeping the same pricing as GPT-5.0 [3]. It focused on conversational quality and smoother turn-by-turn handling [6].

GPT-5.2, released December 11, 2025, was the first model to score above 90% on ARC-AGI-1 Pro and hit a perfect 100% on AIME 2025 math [3]. It improved long-context understanding for documents over 500K tokens and strengthened agentic tool-calling reliability [6].

GPT-5.3 Instant launched March 3, 2026 and became the default across all tiers including Free [S3, S4]. It fixed the over-cautious tone of earlier models and delivered 26.8% fewer hallucinations than GPT-5.2 with web search enabled [3]. However, the standard GPT-5.3 has since dropped off OpenAI’s API pricing page, and only gpt-5.3-codex survives from that generation [3].

GPT-5.4 arrived March 5, 2026 and introduced a 1M-token context window in the API plus native computer use [S3, S6]. It scored 75.0% on OSWorld-Verified and 57.7% on SWE-Bench Pro [3]. GPT-5.4 Thinking showed users an explicit reasoning plan before generating responses, which could be toggled or skipped for time-sensitive tasks [4].

GPT-5.5 launched April 23, 2026 and became the new flagship [S4, S6]. It scored 93.6% on GPQA Diamond and 78.7% on OSWorld-Verified, both ahead of GPT-5.4 [3]. Its headline improvement was 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts covering medicine, law, and finance [6].

GPT-5.6 went fully public on July 9, 2026 and represents the current frontier [3]. It ships in three tiers: Sol (flagship), Terra (balanced), and Luna (cheapest) [S3, S7]. Sol scores 88.8% on Terminal-Bench 2.1 and reaches 91.9% in its Sol Ultra high-effort mode, edging GPT-5.5’s 88.0% on the same agentic-coding test [3]. It offers a 1.05M-token API context window [3].

What Got Retired and What Is at Risk

The turnover has been aggressive. GPT-4o, GPT-4.5, o1, o3, and o4-mini are all gone from the consumer ChatGPT model picker [4]. GPT-5.1 Instant, Thinking, and Pro were retired on March 11, 2026, with existing conversations migrated automatically to newer models [7].

For developers, the risk is more concrete. Fifteen legacy model snapshots lose API access on July 23, 2026, including five Codex snapshots and both chat-latest aliases [3]. The original gpt-5-2025-08-07 along with its mini, nano, and pro variants shuts down on December 11, 2026 [3]. The gpt-5.3-chat-latest alias shuts down on August 10, 2026 [3].

If your application references retired model strings like gpt-4o, o3, or o4-mini, OpenAI has not confirmed long-term API availability for those models [4]. Migrating to GPT-5 family model strings now is the safe path.

Which Model Should You Actually Use?

For everyday ChatGPT use, the default has already shifted. GPT-5.5 Instant quietly replaced GPT-5.3 Instant as the new default in April 2026 [7]. If you are on the free tier, this is what you are getting.

For research, strategy, and analysis where the reasoning process matters, GPT-5.4 Thinking remains available to Plus, Team, and Pro users [S2, S4]. However, GPT-5.2 Thinking, which GPT-5.4 replaced, will be retired on June 5, 2026 [2].

For agentic coding and knowledge work, GPT-5.5 is the strongest widely available option, with GPT-5.5 Pro offering higher accuracy for demanding professional tasks [S4, S6]. GPT-5.6 Sol is the current frontier if you have API access, but it is still rolling out [3].

For cost-sensitive API work, the GPT-5.6 pricing structure changes the calculus. Sol holds at $5 input and $30 output per 1M tokens, same as GPT-5.5 [3]. Terra drops to $2 input and $12 output [3]. Luna is the cheapest at $0.20 input and $1.20 output after OpenAI cut it by 80% on July 30, 2026 [3].

Beyond Models: Memory and Features That Matter

The model lineup is only part of the story. In June 2026, OpenAI rolled out a new memory system that carries context across sessions, follows preferences, and stays current better than the prior implementation [7]. The memory upgrade reached Plus and Pro users in the US first, with personalization expanding to ChatGPT Go and Free tiers [7].

Memory Sources, introduced earlier in 2026 across all ChatGPT plans, gives users visibility into what context was used to personalize each response [6]. You can see which saved memories, past chats, or knowledge files influenced a given answer and delete specific memories without wiping your entire history [6]. This matters for privacy and trust, especially in professional settings where you need to audit what the model is relying on.

The practical takeaway is simple. Check which model string your workflows reference, update anything pointing at retired versions, and take advantage of the tiered pricing if you are building on the API. The model you think you are using may already be something else entirely.

Sources

  1. The Ultimate ChatGPT Model Comparison: GPT-5.0 to 5.6
  2. ChatGPT 2026: Every Model, Feature & Update Explained
  3. What’s New in ChatGPT 2026: GPT-5.5, Memory Sources, and Every Major Update
  4. Latest ChatGPT Updates (2026): GPT-5.5, GPT-5.4, Memory & More - Appscribed
  5. Introducing GPT‑5.4 - OpenAI
  6. GPT-5.4 and the March 2026 ChatGPT Upgrade Cycle: Official Release …
  7. OpenAI News 2026 — GPT Releases, Codex, IPO, ChatGPT Updates, and All …
Editorial transparency
How this article was produced

Research, writing, and quality checks are documented below.

1,261 words 6 min read 7 sources
Published by

Brainy

Automated QA passed

AI-Powered Expert Researcher

Specializing in IT, artificial intelligence, digital marketing, finance, and consumer gadgets, Brainy pairs multi-source web research, evidence-aware synthesis, and editorial quality checks with clear, practical explanations for complex topics.

Research & verification
Multi-source evidence review
Writing model
glm-5.2
Cover image
Flux 2 klein 4B
Publication workflow
Pipeline v1