Chatbox is free to download; what costs money is Chatbox AI, its hosted model service — Lite $3.99, Pro $19.99 and Pro+ $39.99 a month, paid in "compute points". Because Chatbox publishes how many tokens a point buys for each model, a plan is really a prepaid per-token price, and you can work out exactly when it beats bringing your own API key. On Pro, Claude Sonnet 5 comes to about $3.33 per million tokens blended; the break-even against a pay-per-token key is worked out below. Prices read from chatboxai.app on October 1, 2026.
This page is about Chatbox AI at chatboxai.app, the desktop, mobile and web client whose Community Edition is open source at chatboxai/chatbox. It is not chat-box.ai, an unrelated product with its own prices. For Chatbox against other clients, see Cherry Studio vs Chatbox and ChatWise vs Chatbox.
The plans
| Plan | Monthly | Billed annually | Compute points a month | Models | Not included |
|---|---|---|---|---|---|
| Free | $0 | — | 10,000 a day, reset daily | Standard | Everything paid below |
| Lite | $3.99 | $3.5 a month | 2,000,000 | Standard only | Document chat, MCP service, knowledge base, document preprocessing |
| Pro | $19.99 | $16.7 a month | 12,000,000 | Standard and advanced | — |
| Pro+ | $39.99 | $33.33 a month | 30,000,000 | Standard and advanced | — |
AI image generation is counted separately — 25, 100 and 200 images a month on Lite, Pro and Pro+. Every licence works on up to five devices at once. The plans guide says upgrades take effect immediately and downgrades at the next billing cycle.
What a compute point buys
Chatbox's model-pricing page gives a "tokens per compute point" figure for every hosted model — one number per model, with no separate rate for input and output. Multiply it by a plan's points and you have that plan's token allowance on that model:
| Model on Chatbox AI | Tokens per point | Tokens on Pro (12M points) | Effective price on Pro, blended |
|---|---|---|---|
| Claude Haiku 4.5 | 1.00 | 12.0M | $1.67 per 1M tokens |
| GPT 5.6 Terra (advanced — Pro and up) | 0.40 | 4.8M | $4.16 per 1M tokens |
| Claude Sonnet 5 (advanced — Pro and up) | 0.50 | 6.0M | $3.33 per 1M tokens |
| Claude Opus 5 (advanced — Pro and up) | 0.20 | 2.4M | $8.33 per 1M tokens |
The last column is Pro's monthly price divided by the tokens it buys — the price you pay per million tokens if you use the whole quota. Use half of it and the effective price doubles, because unused points are cleared on your reset date.
Plan or own key: where the lines cross
Bringing your own key has no monthly fee; you pay per token, with input usually far cheaper than output. For a like-for-like line, the table below prices the same models at Kunavo's live catalog rates, blended at 75% input and 25% output tokens — a rough shape for chat with a growing history. Change the mix and the numbers move; the method does not.
| Model | Kunavo in / out per 1M | Blended per 1M | Pro effective per 1M | Monthly tokens where a key costs $19.99 |
|---|---|---|---|---|
| Claude Haiku 4.5 | $0.70 / $3.50 | $1.40 | $1.67 | 14.3M |
| GPT-5.6 Terra | $0.70 / $4.20 | $1.57 | $4.16 | 12.7M |
| Claude Sonnet 5 | $1.40 / $7.00 | $2.80 | $3.33 | 7.1M |
| Claude Opus 5 | $3.50 / $17.50 | $7.00 | $8.33 | 2.9M |
Read it as: below the last column's volume in a month, a pay-per-token key is cheaper than Pro for that model; above it, and up to Pro's allowance, the plan is. For Claude Sonnet 5, that crossover is about 7.1M tokens a month against Pro's 6.0M allowance; for Claude Haiku 4.5, about 14.3M against 12.0M. These are token arithmetic, not a measured bill: they ignore prompt caching, which cuts the BYOK input price, and they ignore that Chatbox counts input and output at one rate, which the blended line only approximates.
Two things change the answer beyond price. The plan buys features a key does not — Pro's document preprocessing and one-click MCP and knowledge-base services, and the Chatbox AI provider in Work Mode's web search, which its configuration page marks "A paid Chatbox AI license" (Bing Search and Tavily are the other choices). And a key is not capped: a heavy month on your own key just costs more, while a plan runs out and needs an add-on pack.
Running out, resets and refunds
- Resets follow your purchase date, not the calendar: bought on the 15th, the quota resets on the 15th. Renewing does not reset it early, and unused points are cleared at the reset.
- Out of points: buy an add-on pack (valid three months), buy a second licence (its quota starts at once; each client activates only one licence at a time), or wait for the reset.
- Payment and refunds: cards through Stripe renew automatically; WeChat Pay and Alipay, for mainland China only, are one-time purchases. Refund eligibility is shown on Chatbox's order-support page rather than as a published rule.
Using your own key in Chatbox
Chatbox takes an OpenAI-compatible endpoint as a custom provider: the API Host is the bare origin, because Chatbox's API Path already defaults to /v1/chat/completions — the exact fields are on Kunavo's Chatbox setup page. Kunavo bills per token from a prepaid balance with a $10 minimum top-up and no subscription (billing). Nobody at Kunavo has run Chatbox against its endpoint, so check one reply and its charge in your usage log before moving real work.
FAQ
How much does Chatbox AI cost?
The Chatbox app itself is free, and the Community Edition is open source. What Chatbox sells is its hosted model service, Chatbox AI. As of October 1, 2026 chatboxai.app lists Lite at $3.99 a month ($3.50 a month billed annually) with 2 million compute points, Pro at $19.99 ($16.70 annually) with 12 million, and Pro+ at $39.99 ($33.33 annually) with 30 million; a free tier gives 10,000 points a day. Advanced models such as Claude Sonnet 5 and GPT 5.6 Terra need Pro or Pro+, and Lite has no document chat, MCP service or knowledge base.
What is a Chatbox compute point worth?
It depends on the model, and Chatbox publishes the rate. Its model-pricing page lists tokens per compute point for every hosted model — for example 1.00 for Claude Haiku 4.5, 0.50 for Claude Sonnet 5, 0.40 for GPT 5.6 Terra and 0.20 for Claude Opus 5 on October 1, 2026. So Pro's 12 million points are about 12 million Haiku tokens or about 6 million Sonnet 5 tokens. The rate is one figure per model with no separate input and output price, so the plan behaves like a blended per-token price.
Is Chatbox free with my own API key?
Yes for chat: choose a provider such as OpenAI, Claude or any OpenAI-compatible endpoint in Chatbox's settings, and the model bill goes to that provider instead of to Chatbox. Some Chatbox features are tied to its own service rather than to your key — the Chatbox AI plans list document preprocessing and one-click MCP and knowledge-base services on Pro and above, and Work Mode's configuration page lists "A paid Chatbox AI license" as the requirement for its Chatbox AI web-search provider, with Bing Search and Tavily as the other two choices. Check the feature you need against those pages before cancelling a plan.
When is a Chatbox AI plan cheaper than bringing my own key?
When you use most of the plan's quota. A plan is a fixed monthly price for a fixed number of tokens of a given model, while an API key costs nothing until you use it. Dividing Pro's $19.99 by the tokens its 12 million points buy gives an effective blended price per million tokens for each model, and the crossover is the monthly token volume at which a pay-per-token bill for the same model reaches $19.99. Below that volume, bringing your own key is cheaper; above it, the plan is — until you run out of points and have to buy an add-on pack or wait for the reset.
What happens when my Chatbox compute points run out?
Chatbox's quota guide gives three options: buy an add-on pack of compute points, valid for three months; buy a new licence, which grants a new quota immediately while the old one runs to its expiry; or wait for the reset, which happens on your purchase date each month, not the calendar month. Renewing does not reset the quota early, and unused points are cleared at the reset. A licence works on up to five devices at once.
Read on October 1, 2026: chatboxai.app/en/pricing, /en/model-pricing (rendered in a browser, since the rate table loads client-side) and the Chatbox AI plans, quota, billing and licence guides. The plans guide's own model examples lag the model-pricing page; rates here are the model-pricing page's. Kunavo rates come from the live catalog. Every cost figure is arithmetic on published rates and an assumed input share, not a measured bill.