Exclusive: Plus Users Face Hidden 60K Token Context Limit on ChatGPT 5.2-Thinking

In a startling revelation that challenges OpenAI’s public claims, an independent investigation has uncovered that ChatGPT’s Plus-tier users are subject to a significantly smaller context window than advertised — approximately 60,000 tokens at the user interface, despite the company’s assertion that all paid plans enjoy a 256,000-token capacity. The finding, corroborated by a detailed test conducted by a software engineer known online as @the_koom_machine, suggests a systemic gap between marketing promises and real-world performance, potentially misleading paying customers who rely on long-context interactions for complex tasks such as code analysis, legal document review, and academic research.

The investigation began when the user noticed persistent instances of ChatGPT forgetting earlier inputs within the same conversation, even after providing extensive context. To quantify the issue, the user compiled a lengthy thread involving a multi-file coding project, then asked the model: "What is the earliest message you recall on this thread?" After receiving the model’s response, they copied all subsequent text and pasted it into OpenAI’s AI Studio, which provides accurate token counting. The result: 60,291 tokens. This figure aligns with empirical observations from other users on Reddit and developer forums, who have reported similar truncation behavior — particularly when exceeding 50K–70K tokens.

OpenAI has publicly stated that its "thinking" models, including GPT-4o and its variants, support up to 256K tokens for all paid subscribers. This claim, prominently featured in product documentation and promotional materials, implies that users on the $20/month Plus plan should be able to maintain coherent, multi-turn dialogues spanning hundreds of pages of text. Yet, the empirical evidence contradicts this. The 60K-token ceiling observed is roughly equivalent to 45–50 pages of standard text, or a medium-length novel — far short of the 200+ pages implied by 256K tokens. For users engaged in enterprise-level workflows, such as summarizing legal contracts, analyzing entire codebases, or synthesizing research papers, this limitation renders the model unreliable beyond a fraction of its advertised capacity.

While OpenAI has not officially responded to these findings, the pattern mirrors past controversies involving AI companies overstating model capabilities. In 2023, similar discrepancies were noted with GPT-4’s reasoning performance, leading to widespread user skepticism. This latest revelation risks eroding trust in OpenAI’s transparency, particularly as competitors like Anthropic and Google DeepMind have begun publishing detailed technical benchmarks for their context windows. The absence of official clarification suggests either a technical misconfiguration, a deliberate throttling mechanism, or an undocumented tiering policy — all of which raise ethical concerns about consumer deception in the AI-as-a-service market.

For developers and power users, the implications are profound. Applications built on the assumption of a 256K context window may fail unpredictably in production environments. Automated documentation tools, code assistants, and research summarizers could deliver incomplete or hallucinated outputs without warning. The user who conducted the test recommends that all paying customers perform similar validation tests, especially before integrating ChatGPT into mission-critical workflows. Until OpenAI publishes verifiable, real-time context window metrics — ideally with per-request token logging — users are left to navigate a black box with misleading guarantees.

As AI becomes increasingly embedded in professional and academic workflows, the need for technical honesty from providers has never been greater. This case underscores a broader industry trend: the gap between marketing claims and technical reality. Without standardized disclosure protocols, consumers risk building systems on unstable foundations. OpenAI now faces a critical choice: either clarify its context window policies with full transparency, or risk being perceived not as an innovator, but as an obfuscator.