LLM API price comparison: same model, different provider

Updated 2026-11-19 · Tapix AI

The same model is not the same price depending on where you buy it. Frontier models have a published list price, and resellers, gateways and cloud hosts each charge something slightly different for the identical model name. For the models we carry, Tapix AI prices the same model up to 30% below the vendor's list price, and the two numbers sit side by side on our pricing page so you can check the arithmetic yourself.

That "up to 30%" is not a promotional price. It is how our pricing is built: we buy in volume from upstream, keep our per-token cost below the list price, and pass most of the gap back as a lower per-token rate. You are still calling the same model with the same name and the same capabilities — you are just paying less per token.

Quick answer: the exact same model can cost 30% (or more) less per token from a gateway than from the vendor's own console, because the gateway has negotiated a lower upstream cost and passes part of it through. Always compare on input and output price, never on a single blended number.

The comparison table

Prices below are per 1 million tokens, in US dollars. "List" is the vendor's published price; "Tapix AI" is our price. The numbers are read from our models endpoint on 2026-11-19 and are rounded for display.

ModelVendorList inputList outputTapix AI inputTapix AI outputYou save (output)
claude-opus-5Anthropic$5$25$3.50$17.5030%
claude-sonnet-5Anthropic$2$10$1.40$7.0030%
gpt-5.5OpenAI$5$30$3.50$21.0030%
gpt-6.1-terraOpenAI$1$5$0.43$1.4571%
gemini-3-flashGoogle$0.50$3$0.35$2.1030%
grok-4.5xAI$2$6$1.40$4.2030%

Two things stand out in that table. First, the savings on output tokens is what matters, because output is the line that decides most chat and agent bills. Second, the savings percentage is not uniform — some models carry a sharper discount than others — which is exactly why you compare per model rather than trusting a single headline number.

Why sorting by input price is misleading

Input price is the wrong sort key for one simple reason: you do not pay for most of your input tokens the way you think you do.

A model advertises a low input rate and a higher output rate. If you sort every provider by input price alone, you are optimising the smaller half of the bill. The moment a provider prices input cheap but output high, a table sorted on "input" ranks them on top while quietly costing you more overall. The only number that reflects a real invoice is input cost plus output cost, weighted by how many tokens of each you actually send and receive.

Here is the arithmetic in cost units, so the ratios hold no matter what the current dollar rates are:

A provider whose input price is 10% lower but whose output price is 10% higher wins that input comparison and loses the actual bill. Sort on the total, never on input alone.

Output decides the bill

For the workloads that actually run on an API — chat products, agents, content generation, roleplay frontends — the model generates more than it reads. Generation is priced higher per token than reading across every major vendor, so the two multipliers stack: you produce more output tokens, and each one costs more.

The practical rule: compute your own input-to-output ratio before you optimise anything.

  1. If output is most of your bill, shop on output price first, then check input.
  2. If you push large documents into a long context, input is your dominant line and you should weight input price more heavily.
  3. Either way, compare the same model across providers — a smaller model that shares a name prefix is a different product at a different price.

That last point is the one people get wrong. "Cheaper Claude" can mean a genuinely cheaper rate on the same Claude model, or it can mean a smaller Claude model with a similar name. The difference shows up only when you look at the full model identifier and compare like for like.

A note on how this is possible

The obvious question — "if it is cheaper, is it the same model?" — has a straight answer. We do not swap models. The model name in your request is the model name behind the request; the lower price comes from upstream volume purchasing, not from quietly routing to a smaller model. Our pricing page shows the vendor's list price next to our price for every model, so the substitution question can be settled by reading one table.

What we do not offer is the kind of thing that makes a list price a list price in the first place: formal vendor relationships and the paperwork that comes with them are not something we claim. If your org requires that, buy direct.

FAQ

Q: Is Tapix AI cheaper than the list price on every model?

Not on every model — on the models we carry, our price sits below the vendor's list price by roughly 30%, and the exact figure varies by model. We show both numbers side by side so you can check each one.

Q: Why is the savings percentage different from model to model?

Because our upstream cost differs per model, and per-model demand differs too. A model with a sharper discount simply has a larger gap between our upstream cost and the vendor's list price.

Q: Should I compare on input price or output price?

On both, weighted by your real token mix. Input price alone is misleading because output tokens are usually the larger share of a chat or agent bill.

Q: Is the model I get really the same one?

Yes for the model identifier you request. The price difference comes from upstream volume purchasing, not from swapping in a smaller model. Compare the full model name on both sides to satisfy yourself.

Q: How do I verify these prices instead of trusting this page?

Read the vendor's own pricing page for the list price, then read our pricing page for our number. The two are published side by side on our end, and the vendor's is public on theirs.

See the actual per-model prices. The price page lists the vendor's official price next to ours, row by row, for every model in the live catalog. Open the price list

Related reading

Last updated: 2026-11-21最后更新:2026-11-21