LobeChat pricing now means LobeHub pricing: the software you self-host costs $0, and LobeHub Cloud lists a Free tier plus three paid tiers at $12.9, $24.9 and $49.9 billed monthly, or $118.8, $238.8 and $478.8 billed yearly. The number that decides your real bill is not on either of those lines. It is what a compute credit is worth, because the subscription meters model usage in credits and neither the pricing page nor the docs glossary states a credit-to-dollar rate.
One naming point first, because the search results are full of stale pages. LobeChat is a retired product name. The rename was announced on November 3, 2025, and github.com/lobehub/lobe-chat now redirects to lobehub/lobehub, whose latest stable release is v2.2.18, published September 20, 2026. A separate product called Lobe, at lobe.ai, was a desktop app for training machine-learning models; its GitHub organisation now states that "the Lobe desktop application is no longer under development". It is unrelated, and no price on this page comes from it.
LobeHub Cloud pricing, and the $9.9 that is not a monthly price
Start with the trap, because it produces a wrong budget in one click. lobehub.com/pricing carries a Yearly/Monthly toggle labelled "20% off" and defaults to Yearly, so the plan cards read $9.9, $19.9 and $39.9. Those are per-month-billed-annually figures, computed in the browser from the yearly totals: the only dollar strings in the page source are $0.0, $12.9, $118.8, $24.9, $238.8, $49.9 and $478.8. There is no $9.9 monthly plan to buy.
| LobeHub Cloud tier | Billed monthly | Billed yearly | Credits per month | File storage | Vector entries |
|---|---|---|---|---|---|
| Free | $0.0 | $0.0 | 500,000 | 10.0 MB | 100 |
| Starter | $12.9 | $118.8 | 5,000,000 | 1.0 GB | 5,000 |
| Premium | $24.9 | $238.8 | 15,000,000 | 2.0 GB | 10,000 |
| Ultimate | $49.9 | $478.8 | 35,000,000 | 4.0 GB | 20,000 |
Read from the pricing page source on September 21, 2026. Note that the yearly total for Starter is $118.8 where twelve charges at $12.9 would be $154.80, so the saving is slightly deeper than the toggle label suggests — confirm the figure in checkout rather than trusting either capture. Three further terms the same page states: the pricing FAQ answers "Will the points be reset monthly?" with "Yes, the points will be reset monthly … The points will reset at the end of each billing cycle"; top-ups exist for the paid tiers while Free reads "Subscription required to purchase"; and an Enterprise Edition for private deployment lists a commercial licence, brand theming and a self-hosted provider against a Contact button rather than any published price. Top-up pack sizes and prices are not published either, so neither figure is estimated here. The same FAQ offers a 50% discount code to students and education institutions who write in from an institutional address with proof of status.
What a LobeHub credit is actually worth
LobeHub defines a credit only as a measure. The pricing page says "LobeHub Cloud uses Credits to measure model usage (mapped to tokens)", and the docs glossary says only that Credits are "the unit LobeHub uses to measure AI usage", drawn from "your free allowance, subscription, referral rewards, and top-ups". Neither states a conversion to dollars. But the per-model credit rates are published, and they can be divided.
Take the Anthropic rows of that table and set them against Anthropic's own per-million-token rates (read the same day, legacy rows included): Fable 5.1 and Fable 5 at 10M/50M credits against $10/$50, Opus 5 and Opus 4.8 at 5M/25M against $5/$25, Sonnet 5 at 2M/10M against $2/$10, Sonnet 4.6 at 3M/15M against $3/$15, Haiku 4.5 at 1M/5M against $1/$5. Seven rows, seven exact matches, one divisor: 1,000,000 credits corresponds to US$1.00 of that vendor's published token price. That is derived arithmetic shown so you can redo it, not a rate LobeHub has agreed to. It was checked here against the Anthropic rows only, and it would not survive a repricing: LobeHub could change either side without notice, and rows from vendors that tier their rates by context length will not divide as cleanly. Re-run the division against your own model's published vendor rate before budgeting on it.
| Tier | Credit grant | Token value at the derived rate | Monthly price ÷ that value | Yearly-rate price ÷ that value |
|---|---|---|---|---|
| Free | 500,000 | ~$0.50 | — | — |
| Starter | 5,000,000 | ~$5.00 | 2.58x | 1.98x |
| Premium | 15,000,000 | ~$15.00 | 1.66x | 1.33x |
| Ultimate | 35,000,000 | ~$35.00 | 1.43x | 1.14x |
Do not read those ratios as a margin. The same fee also buys hosting, file storage, vector storage, cloud sync, web search and support, none of which a token rate covers — the ratio measures only how much model value is bundled, and it improves steadily as the tier rises. Ultimate on the yearly rate is the tightest, at close to the token value of what it grants. What the table does show is that a subscription is poor value as a way to buy tokens alone, and reasonable value as a way to buy a hosted product with tokens included.
Does bringing your own key avoid the credit bill?
The plan comparison ticks "Bring your own Provider API keys" on all four tiers, Free included, alongside "Unlimited Message Requests", and the pricing FAQ presents it as the remedy when credits run short: "you can set up custom model API keys to use API usage purchased from other channels." The docs glossary goes one step further under Custom API: "Connecting your own model provider API keys (OpenAI, Anthropic, OpenRouter, and others) instead of using built-in credits. Available on paid plans."
Those two sources do not agree with each other, which is the reason to check before you plan around either. The plan table ticks the row for Free; the glossary says the feature is available on paid plans. And "instead of using built-in credits" is a glossary gloss rather than a billing rule — it is the closest LobeHub comes to saying BYOK traffic consumes no credits, and it is not stated as a term of service anywhere this page could find. Confirm the behaviour on your own tier before you budget on it.
A second boundary matters more for choosing a provider. A key and a base URL are separate permissions, and LobeHub grants them separately.
| Edition | Vendor API key | Custom base URL | What it is documented as |
|---|---|---|---|
| Self-hosted, by environment | Per provider, per deployment | OPENAI_PROXY_URL, ANTHROPIC_PROXY_URL | Environment variables in the canary .env.example |
| In-app, edition not named by the doc | Settings → AI Providers | An optional Proxy URL field, plus a Check button | The User Guide, hedged as "If Proxy URL is available" |
| LobeHub Cloud | Ticked on every tier | Not stated on the pricing page | Unverified here — confirm in your own account |
Counting the environment surface only: the canary .env.example carries ten *_PROXY_URL variables, among them the OpenAI and Anthropic entries, whose commented defaults are https://api.openai.com/v1 and https://api.anthropic.com. It also carries thirty-four *_API_KEY variables. Nine providers have both; Azure and Cloudflare get an endpoint under a different name (AZURE_ENDPOINT, CLOUDFLARE_BASE_URL_OR_ACCOUNT_ID); Ollama has a proxy URL and no key; and twenty-three key variables ship with no endpoint override at all. That count is about the env file, not about LobeHub as a whole — the Settings URL Share schema gives thirteen providers an apiKey and a baseURL in the same object, OpenRouter among them, and OpenRouter has no *_PROXY_URL in the env file. One security note worth carrying from that same doc: it puts the API key in the URL in plaintext, and warns that "LobeHub does not verify the correctness of the settings parameters in the URL, nor provide encryption or decryption methods."
What self-hosting really costs
The software is $0, and the LobeHub Community License — read live on September 21, 2026 — is the thing most write-ups get wrong. It is not plain Apache-2.0. It is "based on Apache License 2.0 with the following additional conditions": you may use it commercially "including as a frontend and backend service without modifying the source code", but "a commercial license must be obtained from the producer if you want to develop and distribute a derivative work". Running it in-house is free; rebranding it and shipping it is the paid line.
Two costs sit behind the $0. The self-hosting docs list "PostgreSQL 14+" as required for a server deployment, and the knowledge-base doc adds the pgvector and pg_search extensions on top of it — that is a database bill and an operator's time, not nothing. And the 2.0 announcement says LobeHub has "decided to officially start exploring commercialization under the self-hosted mode in version 2.0", with "Workspace and API" as the first value-added features aimed at teams and enterprise, alongside a commitment that "all core features will remain permanently open source". Whether any paid self-hosted feature has actually shipped, and at what price, could not be established — treat self-hosted as free to run under the Community License today, with an announced-but-unpriced paid tier on the roadmap.
One capability boundary to settle before you plan a self-hosted knowledge base. Kunavo serves no embedding model, and serves neither text-to-speech nor speech-to-text. A self-hosted knowledge base is configured with its own embedding model — the knowledge-base doc defaults to openai/text-embedding-3-small and lists zhipu, github, openai, bedrock and ollama as the provider options — so that step runs against an outside provider or a local one, separately from whatever answers your chat traffic. On LobeHub Cloud the glossary says LobeHub embeds your files itself, and the vector-entry allowances in the plan table are storage for those results; how Cloud meters that work is not published.
A worked month: credits against catalog rates
This is illustrative token arithmetic, not a measured bill and not a ceiling. Assume one person's month of ordinary LobeHub use: 1,000,000 uncached input tokens — history, agent system prompts and retrieved passages re-sent on every turn — and 60,000 output tokens. Credit figures use LobeHub's published per-model rates; dollar figures use live Kunavo catalog rates for the same token counts.
| Model | LobeHub credits per 1M in / out | Credits this month | Smallest tier whose grant covers it | Kunavo catalog estimate |
|---|---|---|---|---|
| Gemini 3.8 Flash | 0.75M / 3.75M | 975,000 | Starter | $0.68 |
| Claude Haiku 4.5 | 1M / 5M | 1,300,000 | Starter | $0.52 |
| GPT-5.6 Terra | 2M / 12M | 2,720,000 | Starter | $0.95 |
| Claude Sonnet 4.6 | 3M / 15M | 3,900,000 | Starter | $1.56 |
| Claude Opus 5 | 5M / 25M | 6,500,000 | Premium | $2.60 |
Three readings, none of them a verdict. First, the Free tier's 500,000 credits do not cover this month on any of these models — it is a trial allowance, not a working budget. Second, Starter covers four of the five, so at this volume the subscription is buying hosting and storage rather than being out-priced on tokens. Third, the gap opens as usage rises: double the assumption and Starter covers only the two cheapest rows, while the catalog estimate simply doubles. The crossover is your own volume, which is why the honest instruction is to run one month and read the meter rather than to declare a winner.
Kunavo's catalog amount is a billing floor, not 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, image generation and external tools sit outside this example. The minimum top-up is $10 in prepaid credit — a funding minimum, not a task fee or a subscription. See billing details.
Which route wins, and when
| Route | Wins when | What you give up |
|---|---|---|
| LobeHub Cloud subscription | You want hosting, sync, storage and web search handled, and your monthly model usage fits inside a tier's grant | The pricing FAQ says credits reset at the end of each billing cycle, top-up prices are unpublished, and Free cannot reach the frontier rows |
| Direct vendor API | You live in one vendor's models and want its own caching and batch discounts | A second vendor means a second account and a second balance |
| An OpenAI-compatible gateway, self-hosted | You switch models per conversation and want one key and one balance | You run the PostgreSQL instance yourself, and you may have to name the models: the env file takes a per-provider *_MODEL_LIST, and auto-fetching a provider's model list is a per-provider setting rather than a given |
| A gateway on LobeHub Cloud | Potentially the same benefit without self-hosting | Unverified. The plan table promises keys, not endpoints — check it in your own account before planning around it |
| Local model via Ollama | Private or low-stakes work with no per-request charge | A capability gap against hosted frontier models, plus the hardware |
Comparing clients rather than providers? LobeChat vs Open WebUI covers the self-hosting and licence side in detail, LobeChat vs LibreChat the per-user key model, and LibreChat API cost and Open WebUI API cost the same free-client-plus-metered-API split elsewhere. For the provider question on its own, see OpenAI-compatible API and AI cost optimization.
Setting it up
Kunavo publishes a configuration reference for this client. That is a published setup guide, not a compatibility test — LobeHub has not been runtime-tested here against Kunavo's endpoint, and everything above was read from LobeHub's own pricing page, docs, licence, repository and canary sources, plus Anthropic's published rates for the credit arithmetic. Keep a working route available while you try it, run one bounded month, then read what your account actually recorded. Start at the LobeChat and LobeHub integration guide, and create a Kunavo account when you are ready to fund a key.
FAQ
How much does LobeChat cost?
LobeChat is a retired product name — the product is LobeHub, and the repository lobehub/lobe-chat redirects to lobehub/lobehub. There are two prices behind the question. The community edition you self-host costs $0 in software: it is published under the LobeHub Community License, which permits commercial use as a frontend and backend service without modifying the source, and charges only if you develop and distribute a derivative work. LobeHub Cloud is the hosted product, with a Free tier and three paid tiers listed at $12.9, $24.9 and $49.9 billed monthly, or $118.8, $238.8 and $478.8 billed yearly. Prices on a page still headed LobeChat pricing may predate the November 3, 2025 rename, so re-check them against lobehub.com/pricing.
Is LobeHub free?
There is a genuinely free path and a free tier, and they are different things. Self-hosting the community edition has no software cost at all, but LobeHub's self-hosting docs list PostgreSQL 14+ as required for a server deployment, and the knowledge-base doc adds the pgvector and pg_search extensions on top of it — that is infrastructure you pay for, plus whatever your model API costs. LobeHub Cloud's Free tier lists 500,000 credits per month on the plan comparison, and its plan card is headed Monthly Credits Refresh, so it is a recurring allowance rather than a one-time signup bonus. The Free tier cannot spend those credits on the frontier models — fourteen rows on the plan comparison read Not supported in the Free column — and cannot buy credit top-ups, which the table marks Subscription required to purchase.
What is the best API for LobeHub?
It depends on which edition you run, because the two editions do not expose the same door. Self-hosted LobeHub documents per-provider base URL overrides as environment variables: ten _PROXY_URL variables appear in the canary .env.example, and OPENAI_PROXY_URL and ANTHROPIC_PROXY_URL are the two that matter for an OpenAI-compatible or Anthropic-compatible gateway. That env list is not the only door — the Settings URL Share doc also documents a baseURL field per provider, and the User Guide describes an optional Proxy URL field in Settings, AI Providers. On LobeHub Cloud the plan table promises Bring your own Provider API keys on all four tiers including Free, but it never mentions a base URL, and whether Cloud accepts a custom endpoint could not be verified from outside a paid account. So: a direct vendor API wins when you use one vendor all day and want that vendor's own caching and batch discounts. A gateway wins when you switch models per conversation and want one key and one balance, and it is unambiguous on self-hosted LobeHub. The subscription wins when you want the hosting, sync and storage handled and do not want to run PostgreSQL. A local model through Ollama wins for private work with no per-request charge.
What is the cheapest API for LobeHub?
Cheapest listed rate and cheapest way to finish the work are different claims, and in a chat client the gap usually comes from context rather than from the rate card. A chat client re-sends conversation history, agent system prompts and any retrieved knowledge-base passages as input tokens each turn, so a long-running conversation on a cheap model can out-bill a short one on an expensive model. Budget the rate you will pay per million input tokens against the conversation length you actually keep, not against a per-message price. LobeHub publishes its own approximate message counts per plan per model, which are useful for ranking models against each other but rest on an assumed message size it does not disclose.
What is the best model for LobeHub?
There is no single answer, but the shape of the client narrows it. LobeHub is a conversational and agent front end rather than a code-editing agent, so most turns are long-context reads with short answers — which puts the weight of the bill on the input rate. A cheap high-context model handles ordinary chat, summarising and retrieval work; a frontier model earns its rate on reasoning, long-document analysis and agent runs where a wrong answer costs a re-run. LobeHub's own pricing table prices both ends, and its Free tier gates the frontier rows, which tells you where its cost sits. Pick one cheap default and one expensive escalation model, and switch per conversation rather than paying the escalation rate all day.
Does LobeHub Cloud accept a custom base URL, or only a vendor API key?
This could not be established from outside a paid account, and it is the single most important gap on this subject. The evidence for it is circumstantial: the Settings URL Share doc describes keyVaults as provider api key and baseURL settings, gives openai an apiKey and a baseURL in the same object, and writes its worked example against lobehub.com itself. The evidence against treating that as settled is that the doc is filed under self-hosting, the User Guide hedges with If Proxy URL is available, and the plan comparison table promises only Bring your own Provider API keys with no mention of an endpoint. Treat a custom base URL as documented on self-hosted LobeHub, and as unverified on Cloud until you have confirmed it in your own account.
LobeHub's pricing page, licence file, rename announcement, docs glossary, self-hosting and knowledge-base docs and canary environment sources were re-read on September 21, 2026; automated summarising fetches of the pricing page returned wrong figures, so every dollar string here was taken from the page source and from what the plan cards actually render. Anthropic's rates were read the same day. Kunavo token rates come from the live catalog. Cloud base-URL support, top-up pack prices, Enterprise pricing and whether BYOK traffic consumes credits are all marked unverified above rather than estimated.