What Was Rumoured — and What Actually Shipped
The 2025 rumour cycle around GPT-5 coalesced around three claims. First: a 1-million-token context window, up from GPT-4 Turbo's 128K. Second: a summer 2025 launch with an 'extreme reasoning' mode. Third: a step-change in agentic capability that would finally make autonomous tool use reliable.
Here's the 2026 reality. GPT-5 itself launched in August 2025, but with a more modest 400K-token window. The 1M ceiling didn't arrive until GPT-5.4 on 5 March 2026 — eight months late. The 'extreme reasoning' mode became five discrete reasoning-effort levels (minimal, low, medium, high, extreme), now a standard API parameter. Agentic capability shipped as a native Computer Use API that scored 75% on OSWorld — the benchmark that measures whether a model can actually operate a desktop. According to TechCrunch's launch coverage, GPT-5.4 arrived in two variants, Thinking and Pro, with the Pro tier configurable up to the full 1,050,000-token context ceiling.
So the rumour was directionally correct on capability, wildly off on timeline, and underestimated both the pricing and the computer-use breakthrough. If you were a platform company that bet your 2025 roadmap on '1M tokens by Christmas,' you lost a quarter. If you were a marketing team that waited, you got a materially cheaper, materially more accurate model than the rumour promised.
The Community Got the Headline Right, the Timeline Wrong
This is worth dwelling on, because it changes how you should read every 2026 AI rumour. The pattern isn't 'the community hypes things that don't exist.' The pattern is 'the community correctly identifies what's coming but consistently compresses the timeline by 6–9 months.' That's the interesting signal.
Between the July 2025 rumour and the March 2026 launch, three things had to be solved that nobody was talking about publicly: inference-cost economics for million-token calls, latency floors for agent loops, and guardrails for a model with live desktop control. The rumour mill focused on the demo-able capability. The engineering work was in the unglamorous middle. When GPT-5.4 did ship, OpenAI's launch post described responses as 33% less likely to be false compared to GPT-5.2 — an accuracy jump that matters more for regulated marketing content than the context-window headline ever did.
For anyone planning against rumoured AI capabilities — and if you're doing marketing strategy in 2026, you are — add nine months to every whispered roadmap. Budget for the middle.
The 1M Context Window: What 2025 Speculation Missed
In 2025, the rumour coverage focused almost entirely on what a 1M-token window was — roughly 750,000 words, ten full-length novels, the complete content library of a mid-sized B2B brand. Very little of the coverage focused on what it changed.
Here's what the reality turned out to be. Per OpenAI's documentation and the DataCamp technical breakdown, the 272K context window became standard, with 1,050,000 tokens configurable specifically for Codex and enterprise API tiers. The split is deliberate: most workloads don't need a million tokens, and OpenAI priced them accordingly. But when you do need it, three things change that the rumour coverage never predicted.
One: retrieval-augmented generation pipelines built in 2024 are now optional, not required. 'Put the entire brand book in the prompt' became a legitimate architectural choice, not a toy demo. Two: long-running agents stopped needing external memory stores for single-session continuity — the model holds state natively. Three: the cost curve for 'just dump everything in' finally dropped below the engineering cost of building a sophisticated RAG system for 90% of marketing workloads.
That last point is the one the rumour era entirely missed. The interesting question in 2025 was 'can it hold 1M tokens?' The interesting question in 2026 is 'should you bother with RAG at all for anything under 5M tokens of corpus?' Increasingly, the answer is no.
A Marketing Playbook for a Million-Token World
Rather than rehash the capability spec — the cousin post on GPT-5.4's launch covers that — here's what a marketing team should actually put in a 1M-token prompt in Q2 2026.
The brand memory prompt. Load your full brand book, your last 24 months of published content, your three highest-performing campaigns' briefs and post-mortems, and your tone-of-voice rubric. Ask the model to produce a single piece of content that's consistent with all of it. The output isn't marginally better than a 2025 workflow — it's categorically different. Voice drift disappears. Historical contradictions get flagged automatically.
The full-funnel audit prompt. Load every page of your marketing site, the last 90 days of GA4 export, the corresponding ads performance data, and your current-quarter plan. Ask the model 'where are we lying to ourselves?' This is the prompt that GPT-4-era context windows couldn't run, and it's the single highest-leverage analytical use of a million-token window we've seen.
The campaign-replay prompt. Load the complete asset library of a past campaign (briefs, creative, copy variants, performance data, customer support transcripts from the campaign period) and ask the model to rebuild it with a specific variable changed — new audience, new channel mix, new positioning. What used to take a strategist two weeks becomes a two-hour agent loop.
All three are worth running. None of them existed as practical workflows in 2025. Each one is under $10 in API cost per run at current pricing. That economics shift is the playbook — and it's the kind of workflow Anjin was built to run by default.
The Pricing Shock Nobody Predicted
The 2025 rumours had GPT-5 launching at something like $10–15 per million input tokens, based on the GPT-4 Turbo trajectory. The actual GPT-5.4 launch price of $2.50 per million input tokens, documented across NxCode's launch coverage and ApiYi's API guide, reset the flagship market. It's priced below Claude Sonnet 4.6/4.7, in the same neighbourhood as Gemini 3.1 Pro, and it's the single clearest signal that frontier-model quality is being competed down to commodity margins.
Why does that matter for a rumour-retrospective post? Because the rumour era assumed scarcity pricing. The reality is commodity pricing. That assumption shift is more consequential for marketing unit economics than the context window itself. At $2.50/Mtok, 'AI drafts 80% of our marketing' crosses into obviously-profitable territory for any brand spending over £5k/month on content operations. At the rumoured price, it wouldn't have.
What the Rumour Era Taught Us About the Next One
OpenAI is reportedly already briefing partners on GPT-5.5 — codename 'Spud' — and the Information's reporting suggests extreme reasoning modes and further cost reductions. Treat that as rumour, not roadmap. But apply the lessons of the 5.4 cycle: the capability claims will probably be directionally right. The timeline will probably slip by 6–9 months. The price will probably be lower than you expect. And the capability nobody's writing about yet — whatever the 2026 equivalent of 'native Computer Use' turns out to be — will be the one that actually changes how your team works.
The meta-lesson of the rumour era is this. Capability speculation is cheap. Workflow redesign is expensive. The teams that got the most value from GPT-5.4's shipping weren't the ones that predicted the 1M context window — they were the ones that had already designed the prompts that would use it.
Anjin: The Marketing Operating System Built for Million-Token Workloads
Anjin is the Marketing Operating System — a single platform that runs your marketing end-to-end on top of frontier models like GPT-5.4. Not a prompt wrapper. Not a chat interface. An operating system that holds your entire brand context inside the model's working memory, and pairs that context with agents that actually operate your CMS, your ads platforms, your analytics and your distribution.
A million-token context window is what makes that architecture work. Every piece of content Anjin generates is produced against your entire brand history, not a 2,000-word style guide excerpt. Every campaign decision is reasoned about with full performance context loaded. Every agent action is taken with the whole operational picture in view. The three playbook prompts above — brand memory, full-funnel audit, campaign replay — aren't hypotheticals in Anjin. They're default workflows, running against your live data.
Anjin replaces the content agency, the SEO consultant, the paid media planner and most of the coordination software underneath them — and it gets materially better every time OpenAI, Anthropic or Google ships a new model. GPT-5.4 made the platform faster, cheaper and more accurate overnight. Whatever ships next will do the same. We built Anjin for the million-token world that was a rumour in July 2025. It arrived on schedule-minus-nine-months. The platform is already running on it.
The £888 Lifetime License — Offer Closing Soon
Lifetime access to Anjin for a one-time payment of £888. Not a subscription. Not a seat. Not a trial. One payment, unlimited use, for as long as Anjin exists.
The average marketing team spends £888 in about three working days on tooling, freelancers and coordination software. You're buying the platform that replaces most of it — once.
This price will not be offered again once we close our early-access cohort.
Claim your £888 Anjin lifetime license →Founders, agency owners and in-house marketers — this is how you run marketing at AI speed without the team, the burn, or another year of waiting.
Sources: OpenAI — Introducing GPT-5.4, TechCrunch — OpenAI launches GPT-5.4, DataCamp — GPT-5.4: Native Computer Use, 1M Context, NxCode — GPT-5.4 release, pricing 2026, ApiYi — GPT-5.4 API launch guide, The Information — OpenAI's next model, Wikipedia — GPT-5.4
