OpenAI Retires GPT-4.5 From ChatGPT, Closing the GPT-4 Era

GPT-4.5 Leaves ChatGPT on 27 June 2026

OpenAI retired GPT-4.5 from ChatGPT on 27 June 2026, exactly four months after the model launched in February 2026 as the company's declared "most knowledgeable model" at the time of release. The retirement was announced on 28 May 2026 with a 30-day sunset window, giving teams using GPT-4.5 inside custom GPTs time to evaluate the path forward before the model disappeared from the ChatGPT interface. GPT-4.5 is the last of the GPT-4 family to exit ChatGPT, completing a model generation transition that began when GPT-5 arrived and delivered faster responses, lower token costs, and higher benchmark scores than GPT-4.5 at launch.

What GPT-4.5 Was and Why It Mattered

GPT-4.5 launched in February 2026 positioned as OpenAI's most capable general-purpose model at the time, built with an emphasis on depth of knowledge and nuanced reasoning rather than speed or cost efficiency. It occupied the high-capability tier in OpenAI's product line above GPT-4o, making it the default choice for teams that needed the highest-quality outputs and were willing to pay a premium on token cost. During its four months in active service, GPT-4.5 served enterprise customers and individual ChatGPT subscribers who preferred its output style, particularly in research, long-form writing, and complex analysis tasks.

The model's lifespan reflects how quickly the frontier moved in 2026. When GPT-4.5 launched in February, it represented a meaningful step beyond GPT-4o on benchmark tasks requiring wide-ranging knowledge. By the time GPT-5 and GPT-5.3-Codex were established in the market, GPT-4.5 had been superseded on every major benchmark at a higher token cost. The commercial rationale for maintaining a separate GPT-4.5 tier in ChatGPT dissolved once GPT-5 delivered better results at lower cost.

What Replaces GPT-4.5 in ChatGPT

ChatGPT users who were assigned GPT-4.5 as their default model have been automatically migrated to GPT-5 with no action required. Existing conversations that used GPT-4.5 can continue using GPT-5.5, OpenAI's latest and most capable model in the consumer product. GPT-5.3-Codex, which combines the Codex and GPT-5 training stacks with approximately 25 per cent faster inference than its predecessor, is available for technical and coding-focused tasks. For teams that built custom GPTs selecting GPT-4.5 as the underlying model, those configurations now resolve to GPT-5 automatically.

The retirement applies to ChatGPT only. OpenAI has confirmed that GPT-4.5 remains available through the API with no changes to availability or pricing for developers and businesses that have integrated it programmatically. Teams running production systems on the GPT-4.5 API endpoint can continue without migration pressure, though OpenAI's standard model deprecation timeline typically follows a ChatGPT retirement by several months.

The Broader Model Consolidation

GPT-4.5's exit is part of a deliberate consolidation of OpenAI's product model lineup. o3, the reasoning model that launched in April 2026, is scheduled to retire from ChatGPT on 26 August 2026 as o4 and the GPT-5 series reasoning capabilities have overtaken it. OpenAI has also retired or committed to retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini from the ChatGPT interface during this same period. The direction is clear: the GPT-4 generation is leaving consumer products, and the GPT-5 series, including specialised variants for coding and reasoning, is the current standard.

This consolidation reduces user confusion about which model to choose inside ChatGPT while freeing OpenAI's infrastructure and research capacity from maintaining a growing number of production model endpoints. For enterprise customers on API contracts, the migration is a separate conversation with OpenAI's account teams, unaffected by the ChatGPT retirement schedule.

What This Means for Teams Building on OpenAI

For Indian software teams and product companies using ChatGPT for internal workflows, the immediate operational impact is minimal. GPT-5 is a better model than GPT-4.5 on the metrics that matter for most business use cases, and the automatic migration means no workflow reconfiguration is required. Teams using the OpenAI API face no immediate change at all, since the API is unaffected.

For teams building products on the OpenAI API stack, the more relevant signal is where OpenAI's investment is directed. GPT-5.3-Codex is emerging as the production-grade coding model for agentic tasks, offering benchmark improvements over earlier Codex models with faster inference. Teams building AI-assisted code generation, test generation, or documentation tools should evaluate whether their current model selection still reflects the right cost-quality tradeoff now that the GPT-5 series is mature. The 25 per cent inference speed improvement in GPT-5.3-Codex is material for agentic workloads where multiple sequential model calls accumulate latency in user-facing applications.

The model consolidation also signals that the frontier is stabilising at the GPT-5 tier, which is useful for teams making multi-year architecture decisions. Rather than designing for a continuously shifting model landscape, teams can build against a GPT-5 baseline with reasonable confidence that it represents OpenAI's current and near-term standard for production use.

The Bottom Line

OpenAI retired GPT-4.5 from ChatGPT on 27 June 2026, four months after its February 2026 launch as the company's most knowledgeable model. The retirement applies to ChatGPT only; the GPT-4.5 API endpoint is unaffected. ChatGPT users are automatically migrated to GPT-5, which delivers faster, cheaper, and higher-benchmark outputs, making the transition an upgrade rather than a substitution. o3 is scheduled to follow on 26 August 2026. For Indian teams using ChatGPT for internal workflows, no action is needed. For teams building on the OpenAI API, GPT-5 is now the baseline, GPT-5.3-Codex is the model for agentic coding tasks, and the 25 per cent inference speed improvement on the Codex stack is worth evaluating for latency-sensitive applications.