all rankings

ranked on whether you own the mesh, whether you can use it where you live, and whether the geometry is something a game engine will accept — not on how good the render looks.

code-first agent frameworks ranked on production reliability, not github stars — with the durable-execution and low-code layers they sit next to made explicit.

prompt-to-app tools ranked on the two things that actually matter — can you own and export the code, and what the credits really cost — not on how good the first screen looks.

ranked on multi-file changes, codebase understanding, autonomy, editor fit, and what happens when they're wrong.

these vendors bill per 'resolution' and each writes its own definition of the word — so the definition, not the rate, is the price.

ranked on what happens to the voice you clone — who owns it, who may train on it, and whether it's deleted when you leave — then on what a finished minute actually costs.

priced per thousand images, and checked on the question the demos never raise — if the output turns out to infringe something, whose problem is that.

ranked on prompt adherence, text rendering, editing, licence terms you can actually ship on, and real cost per image.

ranked on the questions that involve everyone else in the room — whether the tool announces itself, how long it keeps the recording, and whether your colleagues' voices train somebody's model.

ranked on what the terms let you do with the song afterwards, and on how much legal risk you inherit by using it — the two things the demos never mention.

ranked on what you can actually run today — because a third of this category has been acquired, and the three best options are free.

ranked on motion and prompt adherence, native audio, clip length, licence risk, and cost per second of finished video.

every advertised per-minute rate here covers a different subset of the four things a phone call actually needs — so we priced the whole call instead.

app store optimization tools ranked on data you can trust, pricing you can actually see, and whether an agent can drive them — starting with the free tools most lists skip.

ranked on what they change about your output, what they cost you in context, and whether anyone is still maintaining them.

ranked on whether the skill actually changes what the agent does, how portable it is, and whether the licence lets you use it at work.

the models themselves rather than the assistants built on them, ranked on what the licence file permits and what fits on hardware you own.

ranked on what they'll tell you: what it costs, who actually does the labelling, and what the public record says about how those people are treated.

priced per thousand pages so the numbers actually compare, and checked on the two things that decide it — what a 'page' costs when it has a table on it, and whether your contracts train someone's model.

ranked on accuracy, catch-all handling, price per verification, and how honest they are about what they can't check.

ranked on one question the marketing pages avoid: does the person being monitored know, and can they find out.

ranked on what ten gigabytes actually costs, what 'unlimited' actually means at each vendor, and whether you're told before you pay or after you're abroad.

where to rent an h100 by the hour, ranked on the price you'll actually be billed once egress, storage and idle time are counted — and on whether anyone will promise you uptime.

where to run nous research's hermes agent 24/7 — managed hosts, plain vps boxes and one genuinely free option, ranked on real monthly cost, setup friction, and whether the vendor will still exist next year.

who to buy open-weight model tokens from, ranked on the price of the model you actually want, what they'll admit that price is, and whether they carry the model at all.

ranked on what the licence file says as much as on how well they run — because in this category being called open source and being open source have come apart.

the same database — 2 vcpu, 8gb ram, 100gb storage, 100gb egress, running all month — priced twelve times, because nothing else makes this category comparable.

ranked on what they actually let an agent do, how they authenticate, how alive the project is, and how much blast radius you're accepting.

you buy one of these to stop being the party a tax authority comes after, so we read the clause that decides whether it worked.

ranked on what the licence file actually says, then on capability and what it takes to run — because in 2026 'open source' in a launch post and the terms in the repo are routinely different documents.

ranked on the thing that actually matters for this software — what the gateway is exposed to on first boot — then on price, setup and who patches it afterwards.

ranked on whether anyone outside the company has actually checked the code, how recently, and what happened the last time something went wrong — not on feature checklists.

these tools record what your customers type. we ranked them on what happens when a developer installs the snippet and does nothing else.

ranked on what they need in memory at 4-bit, what the licence permits, and what they're genuinely good at — not on benchmark places they were never going to win.

ranked for developers and agent builders on what you can actually get access to, what it really costs, and whether the api can be driven by an agent at all.

every price converted to the same unit — dollars per hour of audio — plus the two things that change it: whether diarization costs extra, and whether the cheap rate assumes they can train on your recordings.

ai voice models and apis ranked for developers — on price you can read, a licence you can actually ship on, and whether they stream fast enough for a voice agent.

ranked on what sending actually costs at two real volumes, and on how long you can still see the message afterwards — which is the spec nobody compares and everybody eventually needs.

ranked on licence terms you can actually ship on, filtering quality, what it really costs at your scale — and whether you need one at all.

ranked on what they add to meta's per-message rate, how usable the api actually is, and whether they can legally host an ai agent at all.