Why are there so many AI apps?
A new capable model lands almost weekly and every lab wraps its own app around it. Keeping up is a full-time job — which is exactly why you shouldn't have to do it. The churn explained, and the way out.
Why are there so many AI apps?
Because a new capable model lands almost weekly, and every lab wraps its own app around it. Keeping up is a full-time job — which is exactly why you shouldn't have to do it.
A new model every few weeks
The release cadence is real. DeepSeek has shipped open weights continuously since January 2025 — V4-Flash on 31 July 2026, V4-Pro on 13 August 2026. Kimi K3 landed 16 July 2026. GLM-5.3-Flash on 26 August 2026. Qwen 3.8 Flash around 23 August 2026. OpenRouter alone lists about 400 models as of August 2026. The Chinese labs run this cadence because they compete on efficiency, as why Chinese AI models cost less explains; the US labs release less often but aim higher, because they compete on maximum power. Both strategies produce a market where last month's best answer is this month's demo.
Why every lab ships its own app
A model with no interface reaches nobody, so every lab builds one. The interfaces are built by and for technically adept people — the lab's own users. The result is a shelf of apps that assume you know what a context window is and care which checkpoint shipped on Tuesday. Everyone else falls back on the name they have heard of, which is how the default becomes the most expensive option — the mechanics of that are in why is ChatGPT so expensive.
The flaw in the ecosystem
The actual problem is that no good tool just works. The capable cheap models exist. The efficient serving exists. What does not exist, almost anywhere, is a clean app that puts the current best model in front of a person who does not want to think about models. So people default to overpaying, or bounce between free tiers, or give up. The gap is the last mile between the model and the person.
The way out: choose an app
The workable move is to stop shopping for models entirely. Pick one app that maintains the choice for you and re-evaluates it as the frontier moves. That argument, with its full reasoning, is in you don't pick a model, and how Plainly does it — one maintained default, the best available model, no settings sprawl — is in how Plainly handles models. The test of whether this works for writing specifically is in the blind writing test.
The honest limit
If you enjoy tracking the releases — reading the benchmark posts, running each new model through its paces — that is a good hobby. I do it myself, and it is the reason the default in Plainly stays current. But it is a hobby, and it should never be a prerequisite for asking a question. Anyone telling you that you need to compare models before using AI is describing their hobby.
Frequently asked questions
How many AI models are there?+
Roughly 400 are listed on OpenRouter alone as of August 2026, and a new capable open model lands every few weeks. No normal person can track that, and no normal person should have to.
Which AI app should I use?+
One that picks the model for you. The frontier changes every few days, so a maintained default beats a menu.
Do I need a different AI app for different tasks?+
No. Different models have different strengths, but a well-chosen default handles everyday tasks — questions, emails, planning, drafting — without you arbitrating between them.
Why do AI companies release so many models?+
Chinese labs compete on efficiency and ship constantly; US labs compete on maximum capability. Both strategies produce frequent releases. The churn is the market structure, and it will not pass.
Tom Hill · Works at an AI startup
Writes about cheap AI models and honest AI tooling. About the author.
Ready to try Plainly?
A clean, simple place to ask. $5 a month. Unlimited chat. Cancel any time.
Start now