Setup guide · As of November 22, 2026

Cherry Studio Claude API setup in 2.2.3: custom provider and base URL

As of November 22, 2026, the setup in Cherry Studio 2.2.3 is: open Settings from the gear at the bottom left, go to Model Provider, click + Add Provider at the bottom of the list, then Add Custom Provider, give the provider a name and paste the API key that starts with cr_, put https://api.qcode.cc/api in the Anthropic row (no /v1; Cherry builds https://api.qcode.cc/api/v1/messages), click Add and tick claude-sonnet-5-5; for GPT, create a second provider with the same key and put https://api.qcode.cc/openai in the OpenAI row. Tested on Windows 11 (English interface) on 2026-11-21: the reply Hello! came back.

Updated 2026-11-22

Point this tool at the QCode endpoint — one key for Claude, GPT and Chinese models.

#2.2.3 interface#Anthropic row#OpenAI row#one cr_ key

Four values to remember

2.2.3

No provider-type dropdown

In 2.2.3 the Add Custom Provider dialog decides the protocol by which row you fill: Anthropic for Claude, OpenAI for GPT and Chinese models. The dropdown older tutorials describe is gone.

/api

Root address for the Anthropic row

Enter https://api.qcode.cc/api and Cherry builds https://api.qcode.cc/api/v1/messages; the Anthropic preview line on the provider page shows exactly that full path.

/openai

Root address for the OpenAI row

Enter https://api.qcode.cc/openai and Cherry builds https://api.qcode.cc/openai/v1/chat/completions; OpenAI Responses under More options takes the same address and serves GPT only.

cr_

One key, two providers

No Cherry Studio account is needed anywhere in the flow: click Set up later on the welcome page. One key that starts with cr_ covers two providers, one in the Anthropic row and one in the OpenAI row.

How older builds differ from 2.2.3

Adding a provider in Cherry Studio 2.2.3 opens a dialog titled Add Custom Provider with no provider-type dropdown left: the protocol is decided by which Base URL row you fill, so the Anthropic row means the Anthropic protocol and the OpenAI row means the OpenAI protocol. One provider speaks one protocol, so running Claude and GPT side by side means two providers with the same cr_ key.

Claude goes through the Anthropic row only: QCode's OpenAI endpoint answers model_not_available_on_endpoint for Claude models. Older builds call this settings page Model Services; 2.2.3 calls it Model Provider, and it opens from the gear at the bottom left of the app.

Three interface points in 2.2.3

First, Cherry Studio 2.2.3 shipped as the stable release on 2026-11-19 and the English download page reads Stable Release · 11/19/2026. Second, after you click Add, Cherry asks /models under the Base URL for the model list straight away and opens a window titled QCode · Choose models; 2.2.3 has no fetch-model-list button, so refetching later means clicking Sync models on the provider page. Third, the provider page carries a Model Check button for trying a model.

Three dates: the release, the first reply, the second provider

2026-11-19

Cherry Studio 2.2.3 shipped as the stable release, and the English download page writes it as Stable Release · 11/19/2026 · Changelog; that build is the one this setup uses.

2026-11-21

On Windows 11 in the English interface, 2.2.3 with https://api.qcode.cc/api in the Anthropic row, claude-sonnet-5-5 ticked, Say hello in one short sentence sent, Hello! received.

2026-11-21

A second provider named QCode OpenAI added with the same cr_ key, https://api.qcode.cc/openai in the OpenAI row and no typed /v1; after ticking GPT-6.2 Sol the model was listed normally.

Confirmed vs three easy-to-miss points

Confirmed (2026-11-21 hands-on and the docs text)

  1. 1Addresses: the Anthropic row https://api.qcode.cc/api becomes /api/v1/messages, the OpenAI row https://api.qcode.cc/openai becomes /openai/v1/chat/completions, and both take the root address only, with no /v1 and no trailing slash.
  2. 2Models: the Anthropic endpoint returns claude-sonnet-5-5, claude-sonnet-5, claude-opus-5-5 and claude-haiku-5-5; the OpenAI endpoint returns gpt-6.2-sol, gpt-6.2-astra and Chinese models such as glm-5.3.
  3. 3Model list: Add fetches it automatically in a dialog titled QCode · Choose models, later refetches use Sync models on the provider page, and a failure shows Add model manually, a red Failed to pull models line and Retry.
  4. 4Protocol edges: the OpenAI endpoint answers model_not_available_on_endpoint for Claude models, Chinese models also answer on the Anthropic endpoint, and Agent features expect a provider on the Anthropic protocol.
  5. 5Accounts: no Cherry Studio account and no model-vendor account is needed; Set up later on the welcome page is enough.
  6. 6Interface names: 2.2.3 says Model Provider, + Add Provider and Sync models, older builds say Model Services, + Add a provider and sync models.

Three easy-to-miss points

  1. 1Type: 2.2.3 has no type dropdown; putting the root address in the Anthropic row is what selects the Anthropic protocol.
  2. 2/v1: Cherry appends /v1/messages itself, so a typed /v1 turns the path into /api/v1/v1/messages and every request returns 404.
  3. 3Accounts: no Cherry account and no vendor account is needed; click Set up later on the welcome page and use the cr_ key from the console.

Where the API address goes: Anthropic row vs OpenAI row

Anthropic row: Claude, and Chinese models too

Set the Base URL to https://api.qcode.cc/api; the preview path is https://api.qcode.cc/api/v1/messages and the model IDs are claude-sonnet-5-5, claude-sonnet-5, claude-opus-5-5 and claude-haiku-5-5. The docs note that Chinese models also answer on the Anthropic endpoint, and Agent features expect a provider on that protocol.

OpenAI row: GPT and Chinese models

Set the Base URL to https://api.qcode.cc/openai; the preview path is https://api.qcode.cc/openai/v1/chat/completions and the model IDs are gpt-6.2-sol, gpt-6.2-astra and Chinese models such as glm-5.3. Claude is refused on this row, and OpenAI Responses under More options takes the same address and serves GPT only.

Nine steps from the download page to your first reply

  1. 1Open cherryai.com/download, where Windows · x64 is already selected, and click Download Windows Standard; the file is Cherry-Studio-2.2.3-win-x64-setup.exe, about 286 MiB.
  2. 2Run the installer: the English wizard has 4 screens, Choose Installation Options, Choose Install Location, Installing and Completing Cherry Studio Setup. On a clean system it first runs the VC++ runtime installer and raises one user account control prompt from publisher Microsoft Corporation; the app itself is a per-user install with no administrator rights needed.
  3. 3On first launch click Set up later on the welcome page; no sign-in is involved.
  4. 4Click the gear at the bottom left to open Settings, find Model Provider and click + Add Provider at the bottom of the provider list.
  5. 5In Add Custom Provider, type a name such as QCode into Provider Name, paste the cr_ key into API Key, and put https://api.qcode.cc/api in the Anthropic row, root address only, no /v1.
  6. 6Click Add; Cherry fetches models at once and opens QCode · Choose models, so tick Claude Sonnet 5.5 (the grey text on the row is the ID claude-sonnet-5-5) and click Add selected models; a small QCode · Verify and enable window then appears, Cherry sends one test request with the selected model, and you click Done when it finishes.
  7. 7Back in the chat window the model button at the top changes from Select Model to Claude Sonnet 5.5 | QCode; send Say hello in one short sentence and a Hello! reply means the provider works.
  8. 8Optional: click + Add Provider again for a second provider with the same cr_ key, this time putting https://api.qcode.cc/openai in the OpenAI row, again without /v1, and tick GPT-6.2 Sol.
  9. 9To refetch the list later use Sync models on the provider page, or try a model with Model Check; to confirm requests reach QCode, paste the same key at probe.qcode.cc and look at the traffic.

The QCode side: one cr_ key, two providers

The protocol is decided by the request path, not by the key: with one cr_ key, https://api.qcode.cc/api in the Anthropic row reaches Claude and https://api.qcode.cc/openai in the OpenAI row reaches GPT and Chinese models. Swap api.qcode.cc for asia.qcode.cc or us.qcode.cc and both the paths and the key stay as they are.

Billing is per token; current rates for each model are on /models. Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days; if you already have a Personal Plan, add a booster pack when you need more quota. To see whether requests arrive at QCode, paste the same key at probe.qcode.cc.

Cherry Studio setup, step by step

The docs tutorial takes you from the download page to your first reply, with numbered screenshots on Windows in the English interface. The images below show the two finished provider cards: the first has the Anthropic root address and its preview path, the second the OpenAI address with the GPT-6.2 Sol model.

English interface: the QCode provider's Anthropic API Host reads https://api.qcode.cc/api, the preview line reads https://api.qcode.cc/api/v1/messages, and the model list shows Claude Sonnet 5.5.
Check one: the Anthropic API Host reads https://api.qcode.cc/api and the preview line below it reads https://api.qcode.cc/api/v1/messages, English interface.
English interface: the second provider, QCode OpenAI, has https://api.qcode.cc/openai in its API Host (no /v1), and the model list shows GPT-6.2 Sol.
Check two: the second provider's API Host reads https://api.qcode.cc/openai with no /v1, and the model is GPT-6.2 Sol, English interface.

Cherry Studio 2.2.3 on Windows 11, checked on 2026-11-21; the screenshots use the English interface, and wording changes between versions, so follow the version you have.

Common questions

There is no provider-type dropdown in Cherry Studio 2.2.3. How do I pick Anthropic?

In 2.2.3 there is no type to pick: putting https://api.qcode.cc/api into the Anthropic row of the Add Custom Provider dialog is what selects the Anthropic protocol, and Claude models go through that row only. Put them in the OpenAI row instead and QCode's OpenAI endpoint answers model_not_available_on_endpoint. The provider-type dropdown that older tutorials show no longer exists in 2.2.3.

Should the API address in Cherry Studio include /v1?

No. Enter the root address only: https://api.qcode.cc/api in the Anthropic row and https://api.qcode.cc/openai in the OpenAI row, without /v1 and without a trailing slash. Cherry appends /v1/messages or /v1/chat/completions itself, so a typed /v1 produces /api/v1/v1/messages and every request returns 404. The docs add one note: a trailing # turns appending off, which is meant for providers that need a fixed full path, and normal QCode addresses never need it.

What do I do about model_not_available_on_endpoint, or when every request returns 404?

The two point at different mistakes. model_not_available_on_endpoint means the model sits in the wrong row: Claude models were entered under the OpenAI row, which QCode's OpenAI endpoint does not serve, so move the root address to the Anthropic row at https://api.qcode.cc/api. For a 404 on every request, first check whether /v1 was typed into the Base URL (Cherry turns it into /api/v1/v1/messages) and put the Base URL back to the root address https://api.qcode.cc/api; afterwards compare the Anthropic preview line on the provider page, since the path has to match exactly.

Do I need a Cherry account or an Anthropic account to use QCode in Cherry Studio?

Neither: one QCode key that starts with cr_ is all it takes. The docs put it as No Cherry Studio account is needed at any point, and Set up later on the welcome page leads into the app with everything configured locally. One message that looks like a failure but is unrelated: Cherry names conversations with its built-in CherryAI model, so a logged-out app shows Failed to auto-generate conversation name, which has nothing to do with QCode replies.

Why did a single hello use tens of thousands of tokens?

Because the first chat runs in Cherry Support, a pre-built agent that sends a long system prompt with every message, so even one hello can cost tens of thousands of tokens. For everyday chatting click Add Agent at the top, start a plain assistant and pick a QCode model.

After clicking Add I get Failed to pull models, or the model list is empty. What now?

Click Retry in the dialog first; if it fails again, check that the Base URL is the root address (no /v1, no trailing slash), then use Add model manually in the same dialog and type the model ID: claude-sonnet-5-5 for Claude, gpt-6.2-sol for GPT. To refetch the list later, click Sync models on the provider page. To see whether requests actually reach QCode, paste the same key at probe.qcode.cc and look at the traffic.

Sources

QCode docs, Cherry Studio setup tutorial (English, Windows, screenshots updated 2026-11); hands-on run on Windows 11 in the English interface on 2026-11-21; Cherry Studio English download page (2.2.3, stable release 2026-11-19).

Connect both Claude and GPT in Cherry Studio

Create an account for a cr_ key, put https://api.qcode.cc/api in the Anthropic row and https://api.qcode.cc/openai in the OpenAI row; billing is per token, with rates on /models.

Not sure where to start?

Try it for a week

Small Pack: $4 for $30 of quota, valid 7 days

Use it every day

Starter: $9.00/month with $5 of quota every day — just $0.060 per $1 of quota, vs about $0.113 for the Medium Pack