On this page (8 sections)
As of November 2026, Laya Studio’s decision API starts with 100,000 free questions per month, then charges $5 per million questions. There is no per-token metering and no separate output charge, because the model returns a calibrated probability instead of generating text. For high-volume classification, that is usually cheaper than prompting an LLM, and you never get a hallucinated label.
Key takeaways
- This AI decision API pricing comparison 2026 covers Laya Studio, TypeSafe Jev, OpenAI, Google Cloud Natural Language, and AWS Comprehend.
- Laya Studio bills per question: 100,000 free per month, then $5 per million, with no token surcharges.
- OpenAI-style LLM classification bills every input and output token, so costs rise with prompt length and verbosity.
- Ofox tracks per-million-token LLM prices and was last updated August 17, 2026; Price Per Token compares 300+ models.
- Google Cloud Natural Language and AWS Comprehend had no current published rates in the sources I reviewed.
How do AI decision API pricing models work in 2026?
There are three common ways vendors charge for AI decisions. The first is per-token: you pay for every token the model reads and every token it generates. This is how LLM-based classification works. The second is per-request (or per-question): you pay for each classification regardless of how long the prompt is. The third is per-second or per-minute of inference time, common with self-hosted or GPU-backed endpoints.
As of November 2026, Ofox maintains a verified per-million-token price comparison across major LLM providers, and it was updated August 17, 2026 Ofox AI API pricing comparison May 2026. Price Per Token covers 300+ models and is useful for seeing how token costs differ across providers Price Per Token LLM API pricing 2026.
If you are classifying text by asking an LLM “is this a complaint?” you pay for the prompt, the label the model generates, and any explanation text it adds. A short prompt can still get expensive at volume because the output is unbounded. A per-question API caps the cost of each decision up front.
What is Laya Studio pricing?

Laya Studio is a hosted decision API for Laya, the open-source (Apache-2.0) System 1 decision model. You send text or JSON plus typed questions—choice, score, or yes/no—and get a calibrated probability for every question in one forward pass. A calibrated probability is a number that reflects how often the model is right, not a guess wrapped in confident language.
As of November 2026, our pricing is 100,000 free questions per month, then $5 per million questions. There are no input or output token surcharges, and no text is generated, so the model cannot hallucinate a label. We run in Switzerland with zero data retention, and we typically see responses in around 120 ms from Europe. It supports English plus 100+ languages.
Here is what a request and response look like in practice:
This model works well when you know the questions you want to ask in advance. It is not a general chatbot, and it is not designed for open-ended generation. If you need to route a ticket, flag a fraud signal, or score a lead, the per-question price is predictable. For teams concerned about data residency, we wrote about the Swiss data advantage and what it means beyond the EU AI Act The Swiss Data Advantage: Navigating AI Regulations Beyond the EU AI Act in 2026.
What is TypeSafe Jev pricing?

I do not have a current published price sheet for TypeSafe Jev in the sources I reviewed, so I will not quote a per-request or per-token rate here. What I can say is that Laya Studio is compatible with TypeSafe Jev’s /v1/systemone protocol. If you are already sending Jev-shaped requests, switching to Laya means changing your endpoint, not rewriting your request layer.
That compatibility is Jev’s genuine advantage in this comparison: it established a protocol for structured system-one decisions, and existing integrations are a known quantity. The downside, from my perspective, is that I cannot verify its current pricing or data retention terms from the sources available to me today. If you are evaluating Jev, ask for their published rate card and retention policy before you commit.
For teams moving off Jev, we published a step-by-step migration guide How to Switch from TypeSafe Jev to Laya Studio: A Step-by-Step Guide.
What is OpenAI classification API pricing?
When you use OpenAI for classification, you are prompting an LLM and paying for input and output tokens. As of November 2026, Ofox tracks OpenAI’s per-million-token rates alongside other major providers, with its comparison last updated August 17, 2026 Ofox AI API pricing comparison May 2026.
OpenAI’s genuine advantage is flexibility. A well-written prompt can handle ambiguous, nuanced, or multi-label classification that a fixed-schema decision API might miss. The trade-off is cost unpredictability and reliability. You pay for reasoning tokens, politeness tokens, and any text the model generates after the label. You also have to parse free text into a structured decision, and the model can hallucinate a category that does not exist. For low-volume, complex classification, OpenAI can be the right tool. For high-volume, structured routing, the token meter usually works against you.
What are Google Cloud Natural Language and AWS Comprehend pricing?
Neither Ofox, Price Per Token, nor AI Pricing Guru’s published pages in our review listed current per-unit rates for Google Cloud Natural Language or AWS Comprehend, so I will not invent numbers for them AI Pricing Guru API pricing. These are enterprise NLP APIs. Google Cloud Natural Language generally offers entity analysis, sentiment, and syntax. AWS Comprehend offers entity recognition, key phrases, and language detection. Both are typically billed per document or per unit of text, but you should check their current pricing pages directly because I cannot verify a November 2026 rate from the sources provided.
Their genuine advantage is ecosystem integration. If you are already running on Google Cloud or AWS, these APIs sit inside the same console, IAM, and billing account. That reduces operational overhead. The disadvantage is that they are general NLP tools, not calibrated decision engines. You get labels and scores, but not necessarily a calibrated probability that tells you how often the model is right, and you do not get the zero-retention, Swiss-hosted option we offer.
What are the hidden costs beyond the sticker price?
The sticker price is rarely the full cost. With LLM classification, you pay for prompt tokens on every request, and long prompts cost more. You also pay for output tokens, which can balloon if the model decides to explain itself. Then there is the engineering cost of parsing free-text answers into structured labels, and the operational cost of handling hallucinations.
With any API, you should ask about data residency and retention. If you are processing patient messages or support tickets containing personal data, sending that text to a third-party LLM may create compliance obligations. We run in Switzerland with zero data retention, and we have written about what that means for regulated environments Beyond 'Black Box': Implementing Explainable AI with Calibrated Probabilities in Regulated Environments.
A practical rule of thumb I use: estimate your monthly decision volume, multiply by the average prompt plus output token count for an LLM approach, and compare that to a flat per-question price. If the LLM number is higher and your labels are fixed, a dedicated decision API is usually the better value.
Which AI decision API offers the best value for your use case in 2026?
The best value depends on volume and label structure. As of November 2026, Laya Studio’s per-question pricing is predictable for high-volume structured decisions, while OpenAI’s per-token model suits low-volume nuanced work; Google and AWS rates were not published in the sources I reviewed.
| Dimension | Laya Studio | TypeSafe Jev | OpenAI (LLM classification) | Google Cloud Natural Language | AWS Comprehend |
|---|---|---|---|---|---|
| Pricing model | Per question | Not published in sources reviewed | Per token (input + output) | Not published in sources reviewed | Not published in sources reviewed |
| Starting cost (Nov 2026) | $5 per million after 100k free | Not published | Per-million-token rates vary; see Ofox | Not published | Not published |
| Free tier | 100,000 questions/month | Not published | None published | Not published | Not published |
| Output | Calibrated probability (choice, score, yes/no) | Structured decision (via /v1/systemone) | Free-text label | Entities, sentiment, syntax | Entities, key phrases, sentiment |
| Hallucination risk | None (no text generated) | Not published | Possible | Not published | Not published |
| Data residency / retention | Switzerland, zero retention | Not published | Varies by model and region | Not published | Not published |
| Best for | High-volume structured classification, regulated data | Teams already on Jev protocol | Nuanced, low-volume classification | GCP ecosystems needing entity extraction | AWS ecosystems needing managed NLP |
The verdict.
Pick Laya Studio if you are doing high-volume, structured classification—routing tickets, triaging messages, scoring leads—and you want predictable per-question pricing with calibrated probabilities and Swiss data residency. It is the choice I would make for a support or operations pipeline where every decision looks the same and cost must stay flat.
Pick TypeSafe Jev if you are already integrated with its /v1/systemone protocol and do not want to change your request shape. The migration path to Laya exists if you later want per-question pricing and zero retention.
Pick OpenAI only if your classification problem is genuinely ambiguous, low-volume, or requires reasoning that fixed labels cannot capture. Accept that you will pay per token and parse free text.
Pick Google Cloud Natural Language or AWS Comprehend only if you are already committed to that cloud provider and need entity or sentiment analysis alongside classification. Do not pick them because you assume they are cheaper; I cannot verify that from the sources I reviewed, and per-document pricing can surprise you at scale.
If you want to see what a calibrated decision API looks like in practice, check our pricing and docs at Laya Studio.
FAQ
What is the cheapest AI classification API in 2026?
For high-volume structured classification, Laya Studio’s $5 per million questions after a 100,000-question free tier is the lowest verified cost I can quote. LLM-based options bill per token and can cost more once prompts and outputs are counted.
Does Laya Studio charge per token?
No. Laya Studio charges per question. You send text or JSON with typed questions and receive calibrated probabilities without paying for input or output tokens.
Is OpenAI good for classification?
OpenAI is good for nuanced, low-volume classification where labels are not fixed. It is less ideal for high-volume routing because you pay for every input and output token, and the model can hallucinate labels.
Do Google Cloud Natural Language and AWS Comprehend offer calibrated probabilities?
Not in the way Laya Studio does. They return entities, sentiment, and syntax, but I could not verify calibrated probability outputs or zero-retention terms from the sources reviewed.
Can I switch from TypeSafe Jev to Laya Studio?
Yes. Laya Studio is compatible with TypeSafe Jev’s /v1/systemone protocol, so you can change your endpoint without rewriting your request layer. We have a step-by-step migration guide if you want to walk through it.
Topics
- AI decision API pricing comparison 2026
- Laya Studio pricing
- Jev pricing
- OpenAI classification API pricing
- Google Cloud Natural Language pricing
- AWS Comprehend pricing
- cost-effective AI classification APIs
- AI API pricing models
