<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: TapixAi</title>
    <description>The latest articles on DEV Community by TapixAi (@tapixai).</description>
    <link>https://dev.to/tapixai</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4249885%2F7b758762-bd21-488c-80c1-36b160c0025b.png</url>
      <title>DEV Community: TapixAi</title>
      <link>https://dev.to/tapixai</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/tapixai"/>
    <language>en</language>
    <item>
      <title>Why we publish the models where we are more expensive</title>
      <dc:creator>TapixAi</dc:creator>
      <pubDate>Tue, 17 Nov 2026 15:08:02 +0000</pubDate>
      <link>https://dev.to/tapixai/why-we-publish-the-models-where-we-are-more-expensive-3k97</link>
      <guid>https://dev.to/tapixai/why-we-publish-the-models-where-we-are-more-expensive-3k97</guid>
      <description>&lt;p&gt;Every price comparison we publish includes the models where we are &lt;strong&gt;more expensive&lt;/strong&gt; than the official rate.&lt;/p&gt;

&lt;p&gt;That sounds like a strange thing for a reseller to do, so here is the reasoning — and the numbers behind it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The pitch we did not make
&lt;/h2&gt;

&lt;p&gt;We resell API access: one OpenAI-compatible endpoint, many models, prepaid credit. The obvious marketing line is "everything cheaper than list". It is also false, and anyone can check it in five minutes:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;On some models our price sits &lt;strong&gt;below&lt;/strong&gt; the official rate, because of how we buy capacity.&lt;/li&gt;
&lt;li&gt;On others it sits &lt;strong&gt;above&lt;/strong&gt; it, usually because our upstream charges more for that route than the vendor's own direct price.&lt;/li&gt;
&lt;li&gt;On a few it is roughly equal.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The distribution is not stable either. Upstream pricing moves, so a model that is 12% below list today can be 4% above it next month.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why we publish both columns anyway
&lt;/h2&gt;

&lt;p&gt;Three reasons, in order of how much they matter to us:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. The only useful comparison is per model.&lt;/strong&gt;&lt;br&gt;
A single "up to 70% cheaper" claim tells you nothing about your actual bill. If you need one specific model and we are 3% above list on it, the honest answer is "use the vendor directly for that one".&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. A table you cannot audit is marketing, not data.&lt;/strong&gt;&lt;br&gt;
Every row on our pricing page can be checked against two public numbers: our price, and the vendor's list price. When a reader finds a row that does not add up, that is a bug, and we would rather hear it than hide it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. It changes what we build next.&lt;/strong&gt;&lt;br&gt;
The models where we are above list are exactly the ones where our supply agreement is bad. Publishing that number keeps it in front of us instead of burying it in a spreadsheet nobody reads.&lt;/p&gt;

&lt;h2&gt;
  
  
  How the numbers are produced
&lt;/h2&gt;

&lt;p&gt;Roughly every six hours a job measures our real cost per model: it sends a fixed probe, reads the upstream billing counter before and after, and divides by the tokens actually reported. If the sample is too small or the upstream returns garbage, the measurement is marked low-confidence and automatic repricing is frozen for that model instead of guessing.&lt;/p&gt;

&lt;p&gt;Then a pricing floor applies: if the measured cost would push net margin below 20% after fees, the price goes up. Prices never drop automatically — a discount is a decision, not an experiment.&lt;/p&gt;

&lt;h2&gt;
  
  
  What this means if you are comparing providers
&lt;/h2&gt;

&lt;p&gt;Two practical consequences:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Do not trust a comparison table without a "we are more expensive here" column.&lt;/strong&gt; If a reseller only ever shows favourable rows, assume the rest exist and are hidden.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Compare on your own traffic.&lt;/strong&gt; Ten real prompts, one week, two invoices. That beats every table on the internet, including this one.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Our full price list, both columns, is on &lt;a href="https://tapixai.com/pricing" rel="noopener noreferrer"&gt;the pricing page&lt;/a&gt;. If a row looks wrong, email support — we would rather fix it than have it sit there flattering us.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;We run Tapix AI, an OpenAI-compatible API. $1 of free credit on signup, no card required to test.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>api</category>
      <category>llm</category>
      <category>startup</category>
    </item>
    <item>
      <title>What we learned measuring real token costs across 100+ models</title>
      <dc:creator>TapixAi</dc:creator>
      <pubDate>Tue, 17 Nov 2026 12:47:56 +0000</pubDate>
      <link>https://dev.to/tapixai/what-we-learned-measuring-real-token-costs-across-100-models-2l97</link>
      <guid>https://dev.to/tapixai/what-we-learned-measuring-real-token-costs-across-100-models-2l97</guid>
      <description>&lt;p&gt;We run a small OpenAI-compatible API relay. To price anything sanely we had to answer a boring question for every model we carry: &lt;strong&gt;what does one request actually cost us?&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Official price lists did not answer it. So we measured. Over the past two weeks we sent fixed probes to every model on the catalogue and read back the delta on the upstream billing endpoint. A few things came out of that which surprised us, and which apply to anyone building on LLM APIs - even if you never touch a reseller.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. The sticker price is the least interesting number
&lt;/h2&gt;

&lt;p&gt;Every provider quotes two numbers: price per million input tokens and per million output tokens. Your bill is decided by neither of them directly.&lt;/p&gt;

&lt;p&gt;What decided our actual cost, in order:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Output length.&lt;/strong&gt; Output tokens cost roughly 4-8x input tokens. A model advertised at &lt;code&gt;$0.50 / $3.00&lt;/code&gt; is not "a $0.50 model" - the moment answers get long, you are paying the second number.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Retries.&lt;/strong&gt; A 5% failure rate on 100k requests is 5,000 calls you pay for twice, plus the latency.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Prompt size drift.&lt;/strong&gt; A "short" system prompt that quietly grows to 3k tokens changes the bill more than switching between two models in the same tier.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  2. Reasoning tokens are billed even when you cannot see them
&lt;/h2&gt;

&lt;p&gt;This one cost us a week. Some upstream relays return the model's thinking in &lt;code&gt;reasoning_content&lt;/code&gt;. Others return it in &lt;code&gt;reasoning&lt;/code&gt;. Others bill &lt;code&gt;reasoning_tokens&lt;/code&gt; in the usage block and return &lt;em&gt;nothing&lt;/em&gt;.&lt;/p&gt;

&lt;p&gt;If you are building a UI that shows "thinking", you will find that:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;the field name is not standard across relays&lt;/li&gt;
&lt;li&gt;some routes silently drop it while still charging for it&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;usage.completion_tokens_details.reasoning_tokens&lt;/code&gt; is often the only honest signal that reasoning happened&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;We ended up normalising upstream field names into &lt;code&gt;reasoning_content&lt;/code&gt; on our side, and for models whose route does not expose thinking at all, we say so in the model file instead of pretending.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Same model name, different cost
&lt;/h2&gt;

&lt;p&gt;Two relays can serve "the same model" from different account tiers with different prices, and neither is lying. The model is the same; the capacity is not.&lt;/p&gt;

&lt;p&gt;This is why per-model comparison pages are worth less than they look. A comparison table compares one quote against another quote. It does not tell you what the &lt;em&gt;bill&lt;/em&gt; was after retries, caching and output length.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. How we measure, concretely
&lt;/h2&gt;

&lt;p&gt;For each model:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Send a fixed prompt with &lt;code&gt;max_tokens&lt;/code&gt; set high enough that the answer is not truncated (truncated answers make cost look cheaper than it is).&lt;/li&gt;
&lt;li&gt;Read the upstream billing counter before and after.&lt;/li&gt;
&lt;li&gt;Divide by the tokens actually reported - not by the tokens we guessed.&lt;/li&gt;
&lt;li&gt;Repeat until the sample is stable. If it never stabilises, we do not price the model; we mark the cost as low-confidence and freeze automatic repricing for it.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The ugly part: images, video and speech do not bill per token. Per image, per second, per character - each needs its own unit and its own probe. Mixing those into a token comparison produces numbers that are confidently wrong.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. What we do with the numbers now
&lt;/h2&gt;

&lt;p&gt;We keep a 20% net margin floor after fees, and two things enforce it:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;a scheduled job that re-measures cost and raises prices when a model drifts below the floor (it never lowers them automatically)&lt;/li&gt;
&lt;li&gt;a check at process start that walks the whole catalogue, and if any model is priced below the floor it reprices it before serving traffic&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Both write an audit line. Both are boring. Boring is the point: pricing bugs are silent - nothing errors, you just make less money per request until someone notices months later.&lt;/p&gt;

&lt;h2&gt;
  
  
  6. What this means if you are the buyer
&lt;/h2&gt;

&lt;p&gt;Three practical things:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Compare on your own traffic, not on a table.&lt;/strong&gt; Take the ten prompts you actually send, run them against two providers, compare the invoice. Five minutes of work beats any comparison post (including this one).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Watch output length.&lt;/strong&gt; Capping &lt;code&gt;max_tokens&lt;/code&gt; and trimming system prompts is usually a bigger saving than changing vendors.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ask which tier the traffic is served from.&lt;/strong&gt; "Official price, 70% off" is a real thing; so is "same name, different capacity". A provider that can answer that question is usually the one worth keeping.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;We publish our list price next to the official price for every model we carry, including the models where we are &lt;em&gt;not&lt;/em&gt; cheaper - those exist, and hiding them would make the rest of the table worthless.&lt;/p&gt;

&lt;p&gt;If you want to test the numbers yourself, there is a $1 credit on signup at &lt;a href="https://tapixai.com" rel="noopener noreferrer"&gt;tapixai.com&lt;/a&gt;, enough for a few hundred short replies. Break it if you can; that is the useful outcome.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>api</category>
      <category>llm</category>
    </item>
  </channel>
</rss>
