In AI, trust is all you need.

Fuck your benchmarks.
Oh sorry, you didn't hear me?
I said: Fuck. Your. Fake. Ass. Bullshit. Benchmarks.

Can we all just move on from this farce? 

Look, I don't need to one-shot a full Intercom SaaS clone that not a single user ever touches. No one does. Never met em. I need to iterate on a real product for real users. The frontier companies are hyping imaginary benefits to justify their evaporating rationale for burning trillions of dollars on R&D when open source models are proving more effective.

So they allgedly* nerf the real models two weeks before launching the next one so it looks better.

They allgedly* say they can't release "Mythos" because it's "too dangerous". I've heard that before. It was called "GPT-4o". Give me a fuckin break.

The foundation model companies are not on your side.

Silent Nerfs

They degrade models behind the same version number. No changelog. No warning. They think you're too dumb to notice. You just start Googling “did Claude get worse?” and find a thousand other people asking the same thing.

This is not "Open" AI. This is allgedly* fraud-like.

Constant Outages

Major outages every month. Requests queued for minutes at peak. If you're locked to one provider, you're down when they are. Their busy day is your bad day.

Pricing Shell Game

Token costs shift quarterly. Models deprecated with weeks of notice. They train you on the free tier, then jack up prices once you're dependent. Free today, 3x the price by March.

Token Incentives

Verbose responses, bloated context windows. Every token is revenue. Their “thinking” models think longer than they need to because you're paying by the token. Every incentive to make you spend more.

You're the Product

Your prompts train their next model. Enterprise API contracts are different — privacy is paramount — but the consumer tier? You're subsidizing their R&D. Those $20/mo memberships lose money on purpose.

Planned Obsolescence

Every major release follows the same playbook. The current model gets mysteriously worse for a few weeks. Then the new one launches and feels like a revelation. It's not a coincidence. It's weaponized incompetence.

Benchmarks Are Theater

One-shotting an app that works but isn't what you actually need is a pointless metric. Every launch comes with cherry-picked charts. “95% on SWE-bench!” Cool — can it refactor your existing codebase without breaking everything? Can it understand your project's conventions? Can it make something a human actually wants to use?

Benchmarks measure party tricks. Shipping measures everything else.

What the Hell Is Going On

A timeline of broken promises, silent changes, and eroding trust:

Nov 2024 — OpenAI decides they're not a nonprofit anymore. The company built on “safe AI for humanity” announces conversion to for-profit. $150B valuation beats mission statement.

Jan 2025 — GPT-4 Turbo silently degraded. “Did GPT get worse?” trends on Twitter — again. OpenAI says nothing for weeks.

Mar 2025 — Anthropic rate-limits Claude Pro subscribers. $20/mo “unlimited” plans hit invisible usage caps. Users locked out mid-conversation with no warning.

Jun 2025 — Google promises 2M token context. Delivers ~500K. Benchmark says 2 million. Production says “please try again later.”

Sep 2025 — OpenAI deprecates GPT-4 with 60 days notice. Thousands of production apps scramble to migrate. Pricing for the replacement? 2x more.

Feb 2026 — Claude Opus 4.5 launched — then quietly nerfed. Initial release is stellar. Two weeks later, significant quality regression. No changelog. No warning. Just worse.

Mar 2026 — “Mythos is too dangerous to release.” The same thing they said about GPT-4o. Manufacture fear, build hype, delay the launch, drop it anyway. Safety theater as marketing.

Apr 2026 — Silent degradation across all major providers. Rate limits lied about. Access tiers misrepresented. Models swapped behind the same version number. This is normal now.

This isn't a conspiracy. It's a business model. Called grift. Allgedly* of course.

Vera — I named my brand after the latin word for "truth."

I want to build this AI company in the open. and not the OpenAI — we're a nonprofit oh just kidding we're not fuck you — way.

[em dashes are my own thank you very much]. Wait til you meet the en dash for number ranges like 1–3.]

In a world where every AI company has an incentive to mislead you, Vera is built on one principle: the truth. What's working, what's not, and what you should actually use today.

Convention over configuration. The right model is already picked for each task. You don't have to research which one to use.

Full transparency. No black boxes. No hidden routing. No silent swaps. You see exactly which model is running, who made it, and why we picked it.

You're always in control. Swap any default. Bring your own keys. Override anything. The defaults are just a starting point.

One bill. Every model. Stop juggling API keys and credit balances across six providers. One account, one price, access to everything.

Vera isn't a foundation model company. We're not even venture-backed. We have no incentive to lock you into one provider, upsell you on tokens, or train on your data. We just want you to ship — with whatever model is best for the job today.

Vera Studio is free to download. Get started

* My lawyer said to say "allgedly" as much as possible to stop the allegedly* corrupt companies from suing me for slander.