GPT-5.5 and the 'super app': What it means for SMBs

OpenAI shipped a new flagship model yesterday

On April 23, OpenAI released GPT-5.5, the first fully retrained base model since GPT-4.5. The pitch: a smarter, more “intuitive” model that leans into agentic coding, computer use, and multi-step knowledge work. The subtext, delivered by OpenAI president Greg Brockman, is that ChatGPT is moving toward a “super app” — one place where you research, build, execute, and ship, without bouncing between tools.

For a small business owner, the features matter less than the direction. Here is what actually changed, how the economics shook out, and what to do about it.

What GPT-5.5 actually is

Three things stand out from the launch.

It is built for action, not just answers. OpenAI is marketing GPT-5.5 as the first model really optimized for “knowledge work and computer use” — researching online, drafting documents, navigating apps, and staying on a task until it is finished. Mark Chen, OpenAI’s chief research officer, told press the model shows meaningful gains on scientific and technical workflows, including early drug-discovery assistance. The benchmark story is real, but the practical story is that the model is better at running a workflow end-to-end instead of handing you a paragraph and stopping.

It is both cheaper per task and more expensive per token. GPT-5.5 costs $5 per million input tokens and $30 per million output tokens, roughly double GPT-5.4’s headline rate. But the model uses about 40 percent fewer tokens to complete the same work. The net effect is a roughly 20 percent price increase per real-world task, not a 100 percent one. The 1 million-token context window from GPT-5.4 carries over on both standard and Pro variants.

It is rolling out in tiers. Plus, Pro, Business, and Enterprise ChatGPT users get GPT-5.5 now. The Pro variant, which is tuned for harder reasoning and computer-use jobs, is limited to Pro, Business, and Enterprise. Free users stay on older models for now.

Why this matters for small businesses

The easy read is “ChatGPT got smarter again.” The more useful read has three parts.

Model churn is now the baseline

GPT-5.5 arrived just weeks after GPT-5.4, which arrived weeks after 5.2. If you are waiting for the “final” version of the AI you plan to buy, you will be waiting forever. The practical response is to pick tools that upgrade their underlying models automatically and transparently, so you benefit from each new release without re-signing contracts or rebuilding prompts. Vendors that lock you to a specific model version are increasingly a liability.

We covered a related version of this bet in our GPT-5.4 reliability breakdown — the short version is that reliability improvements matter more to a service business than raw benchmarks.

The “super app” direction reshapes how AI reaches SMBs

Brockman’s super app framing is more than marketing. If ChatGPT becomes the default front door for research, shopping, writing, and task execution, the flow of how a small business interacts with software starts to shift. Customers may book, buy, and ask questions inside ChatGPT instead of on your site. Employees may run reports, draft emails, and operate software inside ChatGPT instead of in a CRM.

That does not mean your website goes away. It means the AI layer sitting in front of your customers — the intake widget, the answering agent, the scheduling bot — has to be good enough to compete with a general-purpose super app. If someone can ask ChatGPT “book me an HVAC appointment in Charleston tomorrow,” the HVAC shop whose phone system still sends after-hours calls to voicemail loses the job. That is the core reason we built Hollr and the AI Employees lineup to sit on your channels, not on someone else’s platform.

The cost curve is slowing, but the capability curve is not

Between 2023 and early 2026, per-token prices collapsed. With GPT-5.5, OpenAI is signaling that the per-token price floor is not going to keep falling forever — especially for frontier models. But the model is still completing more work per dollar than GPT-5.4, because it needs fewer tokens. That is a different kind of deflation: you pay the same (or slightly more) per token, and get more done per task.

For your AI budget planning, this means the simple math “model X is cheaper than model Y” is getting less useful. What matters is cost per completed workflow — a chatbot conversation resolved, a content brief generated, a booking completed. Ask vendors for those numbers, not the raw token rate.

Our take

A few things are worth saying plainly.

First, “super app” is a strategy, not a feature. OpenAI has strong incentives to keep users inside ChatGPT. Small businesses should expect that over the next 12 months, more of their customers’ buying journeys will start (and sometimes end) inside a general-purpose AI assistant. That is not a reason to panic. It is a reason to make sure your core channels — phone, web chat, SMS, email — have AI coverage that matches the quality of a GPT-5.5-powered assistant. Otherwise, customers will stay in the super app and never reach you.

Second, agentic coding and computer use are the real GPT-5.5 story. NVIDIA is already running GPT-5.5 inside its internal Codex workflows for engineering work. For an SMB owner, this does not mean you need to hire an AI engineer. It means that the internal tools your vendors build on top of GPT-5.5 — scheduling automation, inventory reconciliation, marketing copy pipelines — should be materially better by summer. Ask vendors what changes after their GPT-5.5 upgrade.

Third, hallucination is still a problem. The Decoder noted GPT-5.5 still hallucinates frequently despite benchmark gains. Do not let a smarter model lull you into removing human review from anything that touches customers, money, or legal commitments.

Bottom line: GPT-5.5 tightens the gap between “AI that answers questions” and “AI that does the work.” Small businesses do not need to use it directly — but the tools you already pay for should get noticeably more capable over the next few weeks, and the bar for customer-facing AI just went up.

What to do this week

1. Ask every AI vendor you pay for a GPT-5.5 roadmap. If your chatbot, scheduling assistant, or content tool runs on OpenAI, ask when they are upgrading and what changes. If they cannot tell you, that itself is a signal.
2. Audit your customer-facing AI against the super app bar. Call your own business after hours. Text your own number. Ask your own chat widget a hard question. If a ChatGPT user would get a better answer inside ChatGPT than from your business, that is a gap worth closing.
3. Do not rewrite your AI stack for GPT-5.5. Model releases this frequent are not a reason to swap tools. They are a reason to pick tools that upgrade themselves — and to re-measure quality and cost per completed workflow once the upgrade lands.

If you want to talk through what the super app era means for your specific business, or how to make sure your intake and scheduling hold up against it, get in touch. We help Appalachian businesses adopt AI in a way that survives the next model release, not just this one.