Msty Studio pricing: the licence, the API bill, and which route wins

Msty Studio's software costs $0 on the Free tier, $149 per user a year for Aurum, $349 once for Aurum Lifetime, and an unpublished quote for Enterprise & Teams — and none of those tiers lists a single model token. No Studio tier bundles inference, so the number worth budgeting is two lines stacked: a licence you may or may not need, and a metered API bill that only starts when you switch a conversation from a local model to a hosted one.

Before any of that, the name. Five different things answer to "Msty", and a price taken from the wrong one is a fabricated price.

That last row matters more than it looks. The legacy v1 documentation at docs.msty.app is still online and still describes a different UI — its own changelog page points readers to the Studio changelog instead. The current Studio documentation lives at docs.msty.ai/studio, and that is the only Msty documentation this page treats as current.

Msty Studio pricing: four tiers, zero included tokens

Read from msty.ai/pricing on September 21, 2026; the per-tier device limits come from the terms, not the pricing page. Three things that page will not tell you directly. First, the current Aurum numbers are the result of a documented increase — lifetime from $249 to $349 and yearly from $129 to $149, posted November 26, 2025 with a December 2 target — so any review still quoting the old pair is stale. Second, the same post states plainly that Msty has never offered discounts and that prices only move up, so do not plan around a sale. Third, lifetime breaks even against the annual licence at about 2.34 years, and its price protection is scoped to the desktop version: the terms say a future web or mobile edition is not covered.

Two windows are worth writing down before you buy: Aurum is fully refundable within 14 days of purchase, and an annual licence can be upgraded to lifetime with a refund of the annual within the first 30 days of the initial purchase (terms, last updated September 2, 2026 and checked September 21, 2026).

The free tier is not free for work

This is the single most expensive thing to get wrong, and it is not in the feature matrix. Msty's terms limit the free licence to "your own, private, non-commercial purposes (e.g. taking notes, doing research)", state that use for a trade or profession you are compensated for does not qualify as Personal Use, and add that work-related activities generating revenue require a purchased licence. The software does not enforce this, so it is a contract limit rather than a gate you will bump into.

The consequence for budgeting: a professional buyer's real Msty total is licence plus tokens, not tokens alone. A hobbyist's total can genuinely be $0 of software plus whatever the hosted calls cost. The same document limits the Msty Go beta to personal, non-commercial use as well, adding that commercial, team or enterprise use may require separate written or paid terms — so "use Go instead" does not route around the question.

What actually bills you: local models versus online providers

Msty Studio is built around local models first, and that is exactly why a long-time Msty user can be surprised by an API bill. Nothing you run locally through Model Hub costs per request. The meter starts the moment a conversation is pointed at an online provider, and from then on that cost sits with whoever issued the key: the documentation describes online providers as using hosted models with your own API keys, and no Studio tier adds an allowance on top.

Studio ships thirteen named online provider types (Anthropic, Cohere, Gemini, Groq, Mistral AI, OpenAI, OpenRouter, Perplexity, Together AI, SambaNova, xAI, Azure OpenAI and Amazon Bedrock) plus a generic bring-your-own path: "If your provider exposes an OpenAI-compatible API, you can connect it by using an OpenAI-compatible endpoint in Model Providers" (online providers documentation, checked September 21, 2026). That generic path is not plan-gated — the matrix marks "Model Hub - Online Models" available on Free, Aurum and Teams alike. The two provider types that are gated are Azure OpenAI and Amazon Bedrock, which need Aurum or Teams.

Three Msty features that move the token bill

Split Chats multiply it. Msty's own documentation says split chats automatically enable duplication mode to sync the message you type between chats. Each pane is its own chat, with its own history to hide or delete, so comparing three online models sends your prompt to three models. Split Chats are available on the free tier, so nothing about your plan slows this down.

Knowledge Stacks cost less than people expect. The documented composing step runs on your own machine: Msty uses gte-tiny by default and tells you to pick replacements from Model Hub > Local Models, where the model needs the Embedding purpose enabled to appear (compose options). Composing a stack that way consumes no paid API tokens. What costs money is the augmented prompt that goes to the chat model afterwards, which is input-heavy by construction. Kunavo serves no embedding model, and none is needed for this route — the documented one keeps that step local. For the general shape of that pipeline see RAG implementation.

On the free tier you cannot see the running total. Insights is the documented in-app view of token usage and cost estimates, and the pricing matrix marks it unavailable on Free; what Free keeps is Response Metrics, marked Limited, which the docs describe as estimated input and output tokens for a single response. Even on Aurum, the Insights documentation warns that "Costs and token usage are estimations only" and tells you to refer to your model provider account for accurate figures (Insights documentation). Treat it as a running gauge, not a ledger.

A worked Msty Studio API cost estimate

These figures are token arithmetic, not measured Msty usage and not a bill ceiling. No one here has run Msty Studio against Kunavo. Assume one long-document conversation that sends 120k uncached input tokens across the session and receives 6k output tokens. The Split Chat column assumes 3 panes of comparable length on the same model, which is an assumption for illustration rather than a measurement. Rates are live Kunavo catalog prices per million tokens.

Three readings. First, the "versus list" column is per model and is not a blanket claim: of the 6 rows above, 5 sit under the model vendor's own list price and 1 sit at it. That unevenness is the honest shape of a catalog, and it is why no flat discount figure appears anywhere on this page. Second, Claude Sonnet 5 and Claude Opus 5 currently carry the same catalog rate, so choosing between those two on this lane is a capability decision rather than a price one. Third, and most usefully for Msty specifically: the Split Chat column is what the comparison habit this product encourages actually costs. If you keep a three-pane split open all day, put one capable model in pane one and cheap models in the rest.

Now put the licence next to the tokens. At the single-conversation estimate above, Aurum's $149 a year is worth roughly 2,483 conversations on Claude Haiku 4.5 but only about 248 on GPT-6 Astra. One Claude Sonnet 5 conversation of that shape estimates at $0.300. For a light personal user the licence is the dominant number and the API bill is rounding; for someone pushing long documents through a flagship all day the ordering flips, and it flips faster the bigger the model. Work out which side of that you are on before deciding whether lifetime is worth $349.

Kunavo's catalog amount is a billing floor rather than a cap: when the upstream reports its charge, the bill is the greater of catalog cost and upstream cost times the applicable markup. Cache charges and external tools sit outside this example, and no prompt-cache saving is claimed here — nothing in Msty's documentation addresses cache headers, so whether Studio passes cache controls through on a custom provider is unknown. The minimum Kunavo top-up is $10 in prepaid credit, a funding minimum rather than a task fee or a subscription; see billing details and AI cost optimization for the measurement method.

Best API for Msty Studio: which route wins when

The provider-native tool presets are the sharpest reason to keep a direct key alongside a gateway. Msty's documentation lists Web Search and Code Interpreter for OpenAI, Web Search, Web Fetch and Code Execution for Anthropic/Claude, URL Context and Google Search with URL Context for Google Gemini, a Responses API with Web Search for xAI, and Browser Search for Groq — noting that availability still depends on the provider, model, account and region. A custom OpenAI-compatible provider is not in that table, so those presets are not yours on the gateway route. If you need them, the honest answer is to keep the vendor key for those conversations. Compare the gateway options themselves in OpenRouter alternatives and best LLM gateway.

Best model for Msty Studio

There is no single answer, because Msty's features pull in different directions. Knowledge Stack answers and long-document chats are input-heavy — the augmented prompt is resent every turn — so the input rate is what moves that bill, and a low-input model like GPT-5.6 Luna or Claude Haiku 4.5 changes the arithmetic more than trimming output ever will. Split Chat comparisons multiply whatever you put in each pane, so the models in panes two and three deserve more thought than the one in pane one. Drafting, summarising and rewriting usually do not need a flagship at all.

Where a flagship genuinely earns its rate is the work where a second attempt is expensive in your time rather than in tokens. That is a judgement about your own tasks, and it is why "cheapest listed price" and "lowest cost to finish the job" are different claims — a cheap model that needs three passes can cost more than one that needs a single pass. For the family-by-family view see Opus vs Sonnet vs Haiku and Claude vs GPT vs Gemini, and cheapest Claude API if you have already settled on one family.

Adding an OpenAI-compatible provider, and what it cannot reach

The documented path is Model Hub > Model Providers > Add Provider, then choose the OpenAI-compatible option and supply your key and endpoint details. Msty's online-providers page stops there; it does not publish the generic form's field labels. The page that does name fields is the Azure OpenAI how-to, which walks through an "API Key" field and a "Base URL" field — for the Azure provider type, not the generic one. So the shape below is a reasonable expectation, not a screenshot of Studio 2.9.6:

Model Hub  >  Model Providers  >  Add Provider  >  OpenAI-compatible

  API key      sk-kn-...
  Base URL     https://api.kunavo.com/v1
  Models       fetch the provider's model list, then pick an id
               (for example claude-sonnet-5)

Two honest caveats on that block. The labels are unconfirmed — the familiar wording "OpenAI compatible endpoint" and the "Fetch Models" button come from the legacy v1 docs and from third-party gateway pages, not from current Studio documentation, and this page has not been checked against an install. The /v1 suffix is a real question with a real answer inside the app: Studio's changelog records release 2.8.3, dated June 23, 2026, "Corrected OpenAI-compatible provider endpoint note wording around the /v1 prefix". That entry says Studio carries its own note about that prefix and that the wording was corrected in June 2026 — read the note in the app and follow it rather than trusting any outside page, this one included. That the code path is live at all is corroborated by release 2.9.5, July 24, 2026: "Fixed OpenAI-compatible provider models and image generation".

Four boundaries a custom provider does not cross in the documented flows. Agent Mode's setup page never selects a Model Hub provider — it installs a CLI agent you sign into with that vendor's account. Vibe CLI Proxy is a browser-authorisation lane for its own list of CLIs, with no key field in its instructions. The provider-native tool presets above are listed per named vendor. And the Azure OpenAI and Amazon Bedrock provider types need Aurum or Teams — the one provider-type plan gate in the matrix, and easy to mistake for a gate on bring-your-own-key generally, which the same matrix shows it is not.

Whether Studio's built-in "Anthropic" provider type exposes an editable base URL is not documented either way, so assume the OpenAI-compatible route unless you check an install. Streaming, tool calling and vision support for the custom provider type are likewise not enumerated in the Studio documentation — which is a gap in the docs, not a statement that they are missing.

Setting it up, and checking the first charge

Kunavo publishes no Msty-specific setup page, and Msty Studio has not been runtime-tested here against Kunavo's endpoint — it is a closed-source desktop app nobody here has installed, and everything above was read from Msty's own site and documentation. What Kunavo does publish is the same OpenAI-compatible shape Msty is asking for: the base URL https://api.kunavo.com/v1 and an sk-kn- key. Cherry Studio is the closest documented parallel — a desktop chat client that takes that same /v1 root and then fetches the model list — and the OpenAI-compatible API guide covers the endpoint itself.

Keep a working route available while you try it, run one bounded conversation, then read the charge your account actually recorded for it. Do not reconcile against Msty's Insights number, which is an estimate by its own documentation and is not shown at all on the free tier. Create a Kunavo account when you are ready to fund a key. If you are still choosing a client, the AI agent API directory lists which desktop and agent clients take a custom endpoint at all.

FAQ

Checked on September 21, 2026, by reading these pages directly: the Msty pricing page and feature matrix, the terms, the Studio and Go changelogs, the price-increase post, the legacy v1 changelog, the Nexus pricing page, and the online-providers, Insights, Responses, Agent Mode, Vibe CLI Proxy, Split Chat, Azure OpenAI how-to and Knowledge Stack compose-options documentation. Kunavo token rates come from the live catalog. Every dollar example here is illustrative token arithmetic under stated assumptions, not a measured Msty session — Msty Studio has not been runtime-tested against Kunavo, and the setup field labels above are marked unconfirmed because no install was checked.