Add every catalog model to Cherry Studio as one provider.

Quick answer: one provider in Model Services.

Click the gear icon in Cherry Studio's left navigation, open the Model Services tab, and click the Add button under the provider list. Give the provider a recognizable name (APIsRouter) and pick OpenAI as the provider type, then save. Now select the new provider in the list and configure it: flip its enable switch, paste your key into the API key field, and set the API address to https://api.apisrouter.com. Note the shape: Cherry Studio's documented default is to take the root address and append /v1/chat/completions itself, so you enter the host without /v1. Then add models with the Add button in the model section, one exact catalog id per entry (claude-sonnet-4-6, deepseek-v4-flash, glm-5.2), and press the Check button next to the key to run a live validation against a selected model.

Provider name:  APIsRouter
Provider type:  OpenAI

then, on the provider page:
API key:        sk-YOUR-APISROUTER-KEY
API address:    https://api.apisrouter.com
                (Cherry appends /v1/chat/completions)
Models → + Add: claude-sonnet-4-6, deepseek-v4-flash, glm-5.2
Check → pick a model → success

How Cherry Studio composes the request URL.

Cherry Studio (CherryHQ on GitHub, roughly 49K stars) is a Chinese-origin desktop client for Windows, macOS, and Linux, known for shipping presets for hundreds of providers and assistants. Anything not in the preset list enters through the custom provider flow above, and the part worth understanding is the API address field. The documented default: you supply the root address, and Cherry Studio concatenates the OpenAI route onto it, so https://api.apisrouter.com becomes https://api.apisrouter.com/v1/chat/completions on the wire. If a provider uses a non-standard path, ending the address with # disables concatenation entirely and the address is used exactly as typed, which is the documented escape hatch. Concatenation behavior around trailing slashes has shifted between releases, so if requests 404, read the final URL in the error rather than guessing; the root-address form plus automatic appending is the stable configuration for a standard /v1 gateway. Once configured, every conversation sends standard chat completions with the model id you added as a plain string. Vendors do not matter to the wire format, which is what lets one provider entry carry Chinese and Western catalog ids together.

Details that earn their keep: multi-key and model management.

The API key field accepts more than one key: separate keys with English commas and Cherry Studio rotates through them top to bottom per request, a documented load-balancing feature. With a gateway that meters per key, this doubles as attribution: one key for work, one for personal, rotation off by listing just one. Model entries are manual and that is a feature. Cherry Studio shows exactly what you add, so the picker stays curated instead of drowning in a fetched catalog. Each entry takes the id as the wire string; the Manage button edits or removes entries later. Group related ids with consistent naming and the picker reads like a menu: fast tier (deepseek-v4-flash, claude-haiku-4-5-20251001), regional strength (glm-5.2, qwen3.7-plus, kimi-k2.6), frontier (claude-sonnet-4-6). The Check button validates the pair of key and address against a model you select, and it is the fastest way to confirm the provider before real conversations. A failed check with a correct key almost always means the address field carries a stray /v1 that doubled the path.

curl -s https://api.apisrouter.com/v1/models \
  -H "Authorization: Bearer $APISROUTER_API_KEY" | head -50
# add these ids verbatim in the provider's model list

Choosing models for a desktop workhorse.

One key means comparisons are picker switches. Run the same week of work on two candidate ids and let the per-model spend in the APIsRouter console referee against your own read of the output quality.

• Daily chat and quick rewrites live on the fast tier: deepseek-v4-flash and claude-haiku-4-5-20251001 keep the always-open window feeling free.
• Chinese-language work is where the regional ids earn their rows: glm-5.2 and qwen3.7-plus for drafting and documents, kimi-k2.6 for long-context reading.
• claude-sonnet-4-6 takes the conversations you forward unedited: careful analysis, client-facing prose, code you intend to ship.
• Cherry Studio's assistants can pin their own model per role, so a translation assistant on glm-5.2 and a coding assistant on claude-sonnet-4-6 coexist behind one provider.
• Add rows deliberately; a curated six-model picker gets used, while a pasted-in catalog gets scrolled.

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

Failure modes specific to Cherry Studio.

The doubled path is the top report: pasting https://api.apisrouter.com/v1 into a field that appends /v1/chat/completions produces /v1/v1 URLs that 404. Enter the root address and let the client build the route. If you must pin an exact URL, the trailing # form exists for that. A failed Check with a good key is usually the address shape above; a failed Check with a clean address is the key, and note that multiple keys must be separated by English commas, since full-width characters pasted from Chinese input methods break the list silently. Models erroring individually while others work is an id-spelling problem; entries are free text and the /v1/models listing is the authoritative source. And remember settings are per machine: a provider configured on your desktop does not exist on your laptop until you repeat the two-minute setup or use Cherry Studio's own backup features. Version drift note: Cherry Studio ships frequently and settings labels move (Model Services has also appeared as Model Provider in some builds); the flow of add provider, choose OpenAI type, key, address, models has stayed stable.

Who routes Cherry Studio through a gateway.

• Bilingual desktop users mixing GLM, Qwen, Kimi, and DeepSeek with Claude and GPT in one picker, one balance, no per-vendor accounts.
• Users in Asia who want Claude and GPT ids on a prepaid balance without a Western card, next to the regional models they already use.
• Power users running Cherry Studio assistants per role who want each assistant pinned to the right vendor without managing five keys.
• People who already route other tools through a gateway and want the desktop client on the same key and usage log.
• Anyone comparing regional and Western models on real daily work rather than benchmarks, where every candidate is one model row.

Verify the endpoint and debug the first conversation.

Curl the models listing first and copy the ids you plan to add; then run one chat completion against your intended daily driver. With both passing, everything left is in the provider page. Inside Cherry Studio, run Check before chatting. Authentication failures are the key field or its comma separators. 404s are the address shape, read the composed URL in the error and count the /v1 segments. Not-found errors on a specific model are id spelling in that row. Once conversations flow, the APIsRouter console shows per-request model, token counts, and spend. A desktop client used across a workday generates a steady stream of small requests, and the per-key usage view turns that stream into per-model, per-day numbers, with Chinese and Western ids on the same page.

curl -s https://api.apisrouter.com/v1/chat/completions \
  -H "Authorization: Bearer $APISROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"deepseek-v4-flash",
       "messages":[{"role":"user","content":"ping"}]}'

FAQ