UPDATED · News · 26 Mar 2026 · MTW News Desk
March 2026 has been a month of dramatic contrasts for OpenAI, marked by the GPT-5.4 Sora shutdown. The company launched GPT-5.4, its most capable model yet with a million-token context window via the API, while simultaneously announcing the discontinuation of Sora, its AI video app that had lost most of its user base after the November 2025 peak. Together, these moves signal a fundamental strategic pivot: OpenAI is done with consumer experiments and is going all-in on enterprise productivity.
What Happened
GPT-5.4: The Million-Token Leap
Released on 5 March, GPT-5.4 arrives in three flavours: a standard version, GPT-5.4 Thinking (optimised for multi-step reasoning), and GPT-5.4 Pro (built for high-performance professional applications). The headline feature is a one-million-token context window via the API, by far the largest OpenAI has ever offered, as OpenAI confirms.
To put that in perspective, a million tokens is roughly 750,000 words, enough to process an entire novel, a complete codebase, or months of business communications in a single prompt. This matters enormously for enterprise users who need to analyse large documents, audit extensive codebases, or process lengthy legal and financial records without losing context, as TechCrunch reports.
The Productivity Pivot
OpenAI’s internal messaging has been blunt. The company is doing away with “side quests” and closing frivolous projects to optimise for productivity on the business front. Sam Altman has stated that ChatGPT must become a productivity tool as the company prepares for a potential IPO at a valuation approaching $1 trillion (around £0.8 trillion), following a record $122 billion funding round that closed in late March at an £675 (about $852) billion valuation, as CNBC reports.
This pivot makes financial sense. Enterprise customers pay significantly more than consumers and are far less likely to churn. With public-market investors expecting OpenAI to price itself above £1 (about $1) trillion at IPO, and the cost base of frontier training still rising, the pressure to demonstrate sustainable revenue growth is immense, according to TechCrunch’s analysis of the Sora wind-down.
The deprecation of GPT-5.1 models (Instant, Thinking, and Pro variants were all removed from ChatGPT on 11 March) further signals this focus, OpenAI is consolidating around fewer, more capable models rather than maintaining a sprawling product line.
What This Means for Users
For ChatGPT subscribers, the immediate impact is positive: GPT-5.4 is a genuine upgrade, particularly for professional work. The expanded context window and improved accuracy make it a more reliable tool for serious tasks. For more, see our AI coverage.
But the Sora shutdown should give users pause. Sora web and app experiences are due to stop working on 26 April 2026, with the API sunsetting on 24 September 2026, according to OpenAI’s own help centre notice. OpenAI has demonstrated that it will kill products, even ones with major partnerships attached, if they don’t align with its evolving strategy. If you’re building workflows around OpenAI’s tools, it’s worth considering whether those tools will still exist in six months. For more, see our news coverage.
The broader lesson is that the AI industry is maturing. The era of flashy demos and viral consumer apps is giving way to the harder, less glamorous work of building reliable enterprise tools. OpenAI’s March 2026 moves suggest the company understands this, even if the transition is messy.
Why OpenAI killed Sora the same month it shipped GPT-5.4
The Sora wind-down landing in the same news cycle as GPT-5.4 is not a coincidence. Sora was one of OpenAI’s most expensive products to run on a per-request basis, and the unit economics never made sense at consumer pricing tiers. Killing it frees up GPU capacity that GPT-5.4’s longer context window and reasoning chains genuinely need, and it removes a product that was steadily dragging gross margin in the wrong direction.
GPT-5.4 itself is the more interesting story. The headline upgrades, a 1-million-token context window, materially better tool use, and a reasoning mode that finally costs roughly what the marketing suggested, are exactly the spec sheet OpenAI needs to fend off Anthropic and Google in the enterprise tier. The model is meaningfully better on long-document tasks and on multi-step agent workflows, which is where the next wave of paying customers actually live.
For end users, the practical implication of GPT-5.4 is that the free ChatGPT tier finally gets a model that is competitive with last year’s paid Plus tier. That is the curve OpenAI has to keep delivering on if it wants to defend its consumer franchise. Killing Sora is the cost of keeping that curve credible.
Buyer action
Where to buy or check next
Use this as the final check before ordering a phone, changing network or trusting a headline monthly price.
Reader discussion
Leave a comment
Comments are moderated. Keep it useful, accurate, and on topic.