OpenAI Begins Phasing Out Legacy Models on ChatGPT Amid Deficits, Enters Cost-Cutting Mode

OpenAI, Retirement of Legacy Models Within ChatGPT
GPT-4o Included Despite Previous Reinstatement After User Backlash
Financial Discipline Sought Through Concentration on GPT-5

OpenAI is set to discontinue support for a large number of legacy artificial intelligence models on ChatGPT, including GPT-4o, which had previously been reinstated following strong user resistance. The move reflects a strategic decision to reallocate computing resources toward the advancement of next-generation models, as the company faces mounting pressure to overhaul its deficit-heavy financial structure and pivot toward a low-cost, high-efficiency operating model.

Official Termination of Legacy Models Including GPT-4o

According to the IT industry on the 1st, OpenAI announced via its official blog that GPT-4o, GPT-4.1, GPT-4.1 Mini, and OpenAI o4-mini will be retired from ChatGPT as of the 13th (local time). These models will be fully removed from the ChatGPT model selection menu starting on the 14th. Previously announced phase-outs of the GPT-5 Instant and GPT-5 Thinking models will also take effect at the same time. Developer-facing API services, however, will remain unaffected by this change and continue operating as before.

Alongside the model consolidation, OpenAI has signaled upcoming changes to ChatGPT’s operating policies. The company plans to reduce unnecessary refusals and admonitory responses, while introducing an “adult-only mode” that grants greater latitude to adult users. To support this initiative, OpenAI intends to deploy age estimation technology globally, while simultaneously strengthening safeguards for minors. The stated objective is to strike a balance between freedom of expression and responsible use.

The retirement of legacy models aligns with OpenAI’s previously stated strategy to concentrate its efforts on GPT-5.2. The company acknowledged that “some Plus and Pro users require more time to transition core use cases, such as creative ideation, to next-generation models.” OpenAI added, “We understand that adaptation takes time, and we recognize that the retirement of GPT-4o may be a significant disappointment for some users. This decision was not made lightly, but it is a necessary step to more intensively improve the models most users rely on.”

GPT-4o, Once Saved by Its Conversational Warmth, Ultimately Phased Out

Initially, OpenAI had planned to remove GPT-4o in August last year to make way for next-generation models, but reversed course within a day after widespread user backlash prompted CEO Sam Altman to intervene personally. GPT-4o, unveiled in May 2024, was the first-generation “Omni” model integrating voice capabilities and played a key role in boosting ChatGPT’s popularity due to its natural and warm conversational style.

Over the past year, however, ChatGPT’s tone has become a recurring source of controversy. While GPT-4o’s friendly and empathetic manner led many users to perceive the AI as more of a companion than a tool, the newer GPT-5 drew criticism for sounding comparatively formal and emotionally detached. Extensive user testing suggested that GPT-4o delivered more expressive and attentive responses, a view echoed by numerous overseas tech publications that published analyses on why GPT-4o outperformed GPT-5 in conversational quality. Debates also flared on Reddit’s ChatGPT forums, with some users disparaging GPT-5 using harsh language.

As disputes over tone intensified, OpenAI responded by restoring GPT-4o as an option for paid users and introducing greater customization of conversational style. Users were given the ability to adjust warmth, enthusiasm, and emoji usage, selecting from options such as default, high, or low. Additional features allowed finer control over tone beyond preset styles introduced in November last year, including Professional, Candid, and Quirky.

Astronomical Deficits Drive Cost Reduction Through Elimination of Token-Intensive Models

Despite its vocal supporters, OpenAI reports that only 0.1% of users engage with GPT-4o on a daily basis, while the vast majority have migrated to GPT-5. The current iteration of GPT-5 has been refined to deliver a warmer and more approachable tone, while fully incorporating the conversational qualities users valued in earlier models and significantly enhancing response accuracy. With internal assessments concluding that GPT-5 now meets both technical and emotional user expectations, OpenAI finalized its decision to retire the legacy lineup.

The move is also driven by the company’s urgent need to escape a high-cost, low-efficiency operating structure. OpenAI remains under heavy financial strain from massive server operating expenses. Altman has previously warned investors that substantial losses could persist for years, underscoring ongoing profitability challenges. According to The Wall Street Journal, OpenAI’s operating losses are projected to reach $74 billion annually by 2028, equivalent to roughly three-quarters of its anticipated revenue.

Legacy models such as GPT-4o have been identified as major contributors to declining profitability due to excessive token consumption during response generation. To sustain the enormous costs associated with data center maintenance and power usage, radical improvements in token efficiency relative to performance have become imperative. GPT-5’s base model is priced at $1.25 per one million input tokens, half the $2.50 charged for GPT-4o. Output token pricing remains at $10 per one million tokens, but enhancements such as automatic recognition of repeated inputs and expanded document processing capacity reduce overall costs for large-scale data workloads. GPT-5 is also more affordable than Anthropic’s Claude Opus 4.1 API, which starts at $15 per one million input tokens and $75 per one million output tokens.

Moreover, GPT-5 is engineered to eliminate redundant computations and derive responses through optimized pathways, maximizing productivity per token. This approach prioritizes qualitative density over sheer response volume, thereby reducing server load. OpenAI’s intensified focus on GPT-5 signals the full-scale activation of its cost-cutting mode. As the operator bearing the heaviest traffic and server expenses in the market, OpenAI occupies a position that virtually compels it to lead the transition toward cost efficiency. Industry experts expect this shift to catalyze parallel evolution among competing models and reshape standards across the AI sector. One AI specialist observed, “From now on, AI profitability will be determined not by how much data a model has been trained on, but by how effectively it delivers the best answers using the least amount of resources.”