Questions and answers
Everything people ask about Zabio, in one place. Something missing? Write to hello@zabio.ai.
- How is Zabio different from other AI routers?
- It picks the model for you, to get you more right answers for less. Send auto and Zabio reads each request, picks the model and reasoning setting that fit the job, and shows why on every response, with what the call cost and what it saved. Pricing stays simple: the provider's list price, a small service fee on each call and a small fixed fee on each auto request. Zabio takes no cut of what routing saves, so every cent of the saving is yours.
- How is Zabio different from OpenRouter?
- OpenRouter is a marketplace with the widest catalog, over 500 models, and it charges 5.5% on every credit purchase whatever you spend it on. Zabio covers fewer models and picks among them for you: it adds a small service fee to each call and a small fixed fee to each auto request, charges nothing on top-ups, and takes no cut of what routing saves you. Both have an automatic router; Zabio's shows the job, the line and the reason on every response.
- How is Zabio different from LiteLLM?
- LiteLLM is an open-source proxy that you host and operate yourself, with your own provider accounts and bills. Zabio is hosted: one key and one prepaid balance for eight vendors, routing that picks the model for each request, teams and quotas included, and nothing to run or patch.
- How is Zabio different from Portkey?
- Portkey, now part of Palo Alto Networks, is an AI gateway focused on observability, guardrails and governance for large companies, and it works with your own provider keys. Zabio focuses on lowering the bill: it picks the model for each request, gives you eight vendors from one prepaid balance, and charges a small service fee per call plus a small fixed fee per auto request, with no cut of what it saves you.
- Can I switch models without losing the conversation?
- Yes. With the Responses API, Zabio stores the conversation and sends the full history to whichever model answers the next turn, on any vendor. When the history outgrows that model's context window, Zabio summarizes the oldest part and continues. With chat completions your app sends the history on every call, so any model gets the full context.
- Can I control what each person and each key spends?
- Yes. Owners and admins set a monthly quota per member, and every API key can have its own monthly spend cap, request and token limits per minute, and a list of the models it may use. Calls stop with 429 quota_exceeded when a limit is reached.
- Does Zabio support prompt caching?
- Yes, on by default where the vendor offers it. For Anthropic, Zabio places cache markers on your system prompt, tools and conversation history, so a long agent loop reads its history from the cache instead of paying for it again on every turn. OpenAI caches automatically. When the turns of a session are more than five minutes apart, Zabio switches to the 1 hour cache so the history is still there. Cached tokens are billed at the vendor's cached rate and shown on every response, and you can set a 5 minute or 1 hour cache per key or per call.
- Does Zabio control reasoning effort?
- Yes. Send reasoning_effort from none to xhigh and Zabio maps it onto the own setting of whichever model answers, on every vendor. With auto and no setting of your own, the cheap and balanced lines use the lowest reasoning setting of the model on easy and medium jobs, so you do not pay for thinking the job does not need. Every response shows the setting used in zabio.reasoning.
- Can Zabio classify text for less than a chat model costs?
- Yes. Name jev-1.13.0, the decision model from TypeSafe AI, and send a JSON schema whose fields are fixed choices, yes or no answers, or bounded scores. Zabio turns each field into one of the typed questions Jev answers and returns the JSON, with a calibrated probability for each field in zabio.decision. Sorting a support ticket into one of five categories with an urgency flag and a 1 to 5 priority cost about 0.00002 USD and took about 0.2 seconds in our tests. Through auto, Zabio uses Jev on its own for such requests whenever it is the cheaper answer; in our tests it got 76 percent of classification items fully right, against 46 percent for the small chat model auto picked before, and answered about five times faster.
- Can I pay less when I am not in a hurry?
- Yes. Send zabio.deadline with a duration of 15 minutes or more, for example "30m", and Zabio serves the call on the flex tier of OpenAI or Google where the model has one, at about half the price. Under auto, candidates are compared at their flex prices. If the vendor has no capacity, Zabio retries at the normal tier and bills the price actually used, so a flex request never fails for that reason. Your saving is measured against the flex price of your usual model, so it never counts the vendor's discount as routing's work. For large jobs that can wait hours, batching sends a whole file at the vendors' batch prices.
- How does batching work on Zabio?
- Upload a file of chat requests in the same format as OpenAI's Batch API and create a batch. Zabio routes every line like a normal call, at the vendors' batch prices, about half the list price at OpenAI, Anthropic, Google, Mistral and Together AI, and returns the answers within 24 hours, often much sooner. Inside a batch, auto only picks models with a batch tier; a line that names a model without one stops the batch before anything runs or is billed.
- What is Smart Splitting?
- Under auto, a request that carries a long document and asks about only part of it can be split: a cheap model copies out the lines the question needs, word for word, Jev checks that the extract holds everything the answer needs, and the model auto picked answers from that extract instead of the whole document. If the check fails, Zabio sends the full document and pays for the attempt itself, so a split never costs you more. In our tests it cut the cost of long-document questions by about a third on average, and by more than half on contracts, with every fact kept. It never applies to a model you name, every step shows in zabio.split, and zabio.split set to off turns it off for a request.
- Is Zabio free?
- Yes. Every account starts on the Free plan: 100 requests a day, up to 16,000 tokens in and 2,000 out per request, on six models from OpenAI, Google, Mistral and Together AI, with auto choosing among them. No card, no expiry. Every other model stays visible, one top-up away, and a daily fair-use limit keeps Free sustainable. Your first top-up removes every limit. Calls on your own vendor keys pay no service fee.
- What fees does Zabio charge?
- Two small ones, and never a cut of your saving. Every call pays a service fee, a percentage of the provider price, which covers what it costs us to serve it: infrastructure, payment processing and support. Every auto request also pays a small fixed fee for automatic model choice, once per request, which pays for reading the job and picking the model. Calls on your own vendor keys pay no service fee, and their auto requests pay the fixed fee. Every cent routing saves stays with you. The current numbers are on the pricing page, and a fee that goes up is announced at least 30 days ahead by email to every organization owner.
- Do I need to change my code?
- Two things: the base URL and the API key. Zabio speaks the OpenAI API, so the official SDKs for Python and Node work unchanged. Zabio specific options, such as session ids or a vendor list, travel in an extra zabio field.
- Which models can auto pick?
- Any chat model enabled for your key, across every vendor Zabio serves. Name a line with auto/best, auto/balanced or auto/cheap, or send plain auto and Zabio chooses for the job: the cheapest model on easy jobs, the best line when the stakes are high, and balanced otherwise. Frontier models are candidates on best only. To narrow the field, switch vendors or models off, or give the key an allow list.
- What happens if the model I asked for is down?
- When you name a model, Zabio retries on another model of the same vendor and class by default. You can ask for a cross vendor fallback or no fallback at all. With auto, Zabio moves to the next model its rule picks. The response always tells you which model actually answered.
- What happens when my balance reaches zero?
- New calls are refused with 402 insufficient_credits before any provider work is done. A model you name that runs on your own vendor key keeps working; auto needs credit, because Zabio bills its small fixed fee for picking the model. Add credits and requests resume at once. Owners and admins get an email at 20%, 5% and zero.
- Do you store my prompts?
- Not by default. Zabio stores call metadata such as model, token counts and cost. Prompts and completions are stored only when you turn on body logging for a key or when you use stored conversations, for a retention period you control. Your content is never used for training.
- How does the Zabio launch match work?
- Spend at least 250 USD in the 30 days after you sign up, and from then on Zabio matches every top-up dollar for dollar until it has added 250 USD. Add 20, we add 20; add 50, we add 50. The match is one per account, in your first organization. The credit joins your balance, works on any call and never expires.
- How does the Zabio referral program work?
- Every Zabio user has a referral link. When a new developer who signed up through it adds 100 USD in top-ups, the person who invited them gets 25 USD in credit once those top-ups are 30 days old. Every referral counts, and the credit never expires.
- What can promotional credit pay for?
- Any call. Credit from the launch match and from referrals joins your balance like a top-up, never expires and has no cash value.
- Can I refund a matched top-up?
- No. A top-up that earned a match cannot be refunded, and if it is disputed, the match it earned is removed.
Build your prototype free
Change two lines in your code, send model auto, and see what every call cost and what you saved.