--- title: "LiteLLM setup: one OpenAI-compatible key for every model" canonical: https://uplinkswitch.ai/guides/litellm generated: 2026-11-04T07:21:57.760Z source: https://uplinkswitch.ai/guides/litellm base_url: https://api.uplinkswitch.com/v1 --- [›uplinkswitch](https://uplinkswitch.ai/) [api.uplinkswitch.com/v1](https://uplinkswitch.ai/models)[Console](https://uplinkswitch.ai/console) [›uplinkswitch](https://uplinkswitch.ai/) # LiteLLM One key, every model: model openai/ with api\_base in the SDK, or a config.yaml model\_list for the proxy. Pay per token, no subscription. Checked 2026-10-28 ·[All guides](https://uplinkswitch.ai/guides)·[Models](https://uplinkswitch.ai/models) At the end you have LiteLLM, the SDK or the proxy, calling `https://api.uplinkswitch.com/v1` with one API key from the console. One key, every one of the 337 chat models on the [list](https://uplinkswitch.ai/models), billed per token from prepaid credit with no subscription. The config below is copied from LiteLLM's own docs (read 2026-10-28); the price table is live. ## Before you start You need Python 3.10 or newer, an account on the [console](https://uplinkswitch.ai/console) with credit on it, and one key created there. Create the key for LiteLLM alone, so its requests sit apart from any other tool on the Usage page. Nothing is installed on the UplinkSwitch side: the base URL and the key are the whole configuration. **Base URL**: `https://api.uplinkswitch.com/v1`. Keep the `/v1` on the end; LiteLLM adds `/chat/completions` itself. ## The SDK: openai/ with api\_base LiteLLM treats any model named `openai/` as an OpenAI-compatible endpoint and sends it to `api_base` with `api_key` (docs, read 2026-10-28). The `openai/` prefix is LiteLLM's routing word, not the developer of the model: `openai/claude-sonnet-5-5` is Claude, reached over the OpenAI format at this base URL. shell ``` pip install litellm ``` litellm\_call.py ``` import os import litellm response = litellm.completion( model="openai/claude-sonnet-5-5", # openai/ prefix = OpenAI-compatible endpoint api_base="https://api.uplinkswitch.com/v1", api_key=os.environ["UPLINKSWITCH_API_KEY"], extra_headers={ "HTTP-Referer": "https://uplinkswitch.ai", "X-Title": "LiteLLM", }, messages=[{"role": "user", "content": "Say ok."}], ) print(response.choices[0].message.content) ``` `extra_headers` is LiteLLM's parameter for custom request headers. `HTTP-Referer` and `X-Title` name the client in the gateway's request log; they change nothing about the answer. `base_url` is an alias for `api_base` if you prefer the OpenAI spelling. ## The proxy: config.yaml model\_list The proxy reads a `config.yaml`. Each entry in `model_list` has a `model_name`, the alias your clients send, and `litellm_params` with the real target: `model`, `api_base` and `api_key`. `os.environ/UPLINKSWITCH_API_KEY` reads the key from the environment so it never sits in the file. shell ``` uv tool install 'litellm[proxy]' ``` config.yaml ``` model_list: - model_name: claude-sonnet-5-5 litellm_params: model: openai/claude-sonnet-5-5 api_base: https://api.uplinkswitch.com/v1 api_key: os.environ/UPLINKSWITCH_API_KEY extra_headers: {"HTTP-Referer": "https://uplinkswitch.ai", "X-Title": "LiteLLM"} - model_name: claude-opus-5-5 litellm_params: model: openai/claude-opus-5-5 api_base: https://api.uplinkswitch.com/v1 api_key: os.environ/UPLINKSWITCH_API_KEY extra_headers: {"HTTP-Referer": "https://uplinkswitch.ai", "X-Title": "LiteLLM"} - model_name: gpt-6.1-sol litellm_params: model: openai/gpt-6.1-sol api_base: https://api.uplinkswitch.com/v1 api_key: os.environ/UPLINKSWITCH_API_KEY extra_headers: {"HTTP-Referer": "https://uplinkswitch.ai", "X-Title": "LiteLLM"} - model_name: gemini-3.1-pro-preview litellm_params: model: openai/gemini-3.1-pro-preview api_base: https://api.uplinkswitch.com/v1 api_key: os.environ/UPLINKSWITCH_API_KEY extra_headers: {"HTTP-Referer": "https://uplinkswitch.ai", "X-Title": "LiteLLM"} - model_name: deepseek-v4-pro-0813 litellm_params: model: openai/deepseek-v4-pro-0813 api_base: https://api.uplinkswitch.com/v1 api_key: os.environ/UPLINKSWITCH_API_KEY extra_headers: {"HTTP-Referer": "https://uplinkswitch.ai", "X-Title": "LiteLLM"} - model_name: kimi-k3 litellm_params: model: openai/kimi-k3 api_base: https://api.uplinkswitch.com/v1 api_key: os.environ/UPLINKSWITCH_API_KEY extra_headers: {"HTTP-Referer": "https://uplinkswitch.ai", "X-Title": "LiteLLM"} litellm_settings: drop_params: true ``` The aliases match the ids on the Models page on purpose, so a client that already knows `claude-sonnet-5-5` needs no change. `drop_params` makes the proxy drop an OpenAI parameter a model does not take instead of raising. Start it and point any OpenAI client at port 4000: shell ``` export UPLINKSWITCH_API_KEY="sk-uplink-live-..." litellm --config config.yaml # Proxy running on http://0.0.0.0:4000 ``` client.py ``` import openai client = openai.OpenAI(api_key="anything", base_url="http://0.0.0.0:4000") response = client.chat.completions.create( model="claude-sonnet-5-5", # the model_name alias from config.yaml messages=[{"role": "user", "content": "Say ok."}], ) print(response.choices[0].message.content) ``` The `api_key` in the client is whatever the proxy expects (anything, until you set a `master_key` under `general_settings`). The UplinkSwitch key stays on the proxy side. ## Which model ids to use Any chat id on the [Models page](https://uplinkswitch.ai/models), with `openai/` in front for LiteLLM. Tool calling and streaming work on every chat model here, so an agent framework on top of LiteLLM can use any of them. The 6 in the config above cover the usual coding and agent picks: `claude-sonnet-5-5` and `claude-opus-5-5` for planning and hard edits, `gpt-6.1-sol` and `gemini-3.1-pro-preview` as the other flagships, `deepseek-v4-pro-0813` and `kimi-k3` where input volume is high. The [uncensored builds](https://uplinkswitch.ai/uncensored) take the same form: one more `model_list` entry with their id after `openai/`. Copy ids from the Models page rather than typing them; a dot or a date suffix out of place is a 404. ## Price per token | Model | Input / 1M | Output / 1M | Cached input / 1M | List input | List output | Under list | | --- | --- | --- | --- | --- | --- | --- | | [Claude Sonnet 5.5](https://uplinkswitch.ai/models/claude-sonnet-5-5) | $1.20 | $3.00 | $0.075 | $2.00 | $10.00 | 40% on input | | [Claude Opus 5.5](https://uplinkswitch.ai/models/claude-opus-5-5) | $2.80 | $10.00 | $0.15 | $4.00 | $20.00 | 30% on input | | [GPT-6.1 Sol](https://uplinkswitch.ai/models/gpt-6.1-sol) | $0.60 | $3.00 | $0.06 | $2.00 | $10.00 | 70% on input | | [Gemini 3.1 Pro Preview](https://uplinkswitch.ai/models/gemini-3.1-pro-preview) | $1.20 | $7.20 | $0.12 | $2.00 | $12.00 | 40% on input | | [DeepSeek V4 Pro 0813](https://uplinkswitch.ai/models/deepseek-v4-pro-0813) | $0.792 | $1.188 | $0.0625 | $1.32 | $3.96 | 40% on input | | [Kimi K3](https://uplinkswitch.ai/models/kimi-k3) | $0.90 | $4.50 | $0.12 | $3.00 | $15.00 | 70% on input | USD per million tokens, live from GET /v1/catalog. List: the developer’s own published price. Cached input bills at the cache-read rate where the model has one; LiteLLM needs no switch for it. LiteLLM's own cost tracking uses its bundled price map, which knows the developers' list prices and not this base URL. For the real charge read the Usage page, or `x-uplink-charge` on a non-streamed response. Every price on this page is held for 30 days from each top-up. ## Check it Confirm the key before the first LiteLLM call: shell ``` curl https://api.uplinkswitch.com/v1/models -H "Authorization: Bearer $UPLINKSWITCH_API_KEY" ``` Then run the SDK script, or start the proxy and run the client. The request shows on the [console](https://uplinkswitch.ai/console) Usage page within seconds, with its tokens and its charge. ## Errors you will see - `401 invalid_api_key`: the key is wrong or revoked, or the variable behind `os.environ/UPLINKSWITCH_API_KEY` was not exported in the shell that started the proxy. Copy the key again from the console and export it. - `402 insufficient_balance`: the balance does not cover the request. Top up in the console. - `402 monthly_limit_reached`: the key hit the monthly spend limit you set on it. Raise it in the console, or wait for the first of the month. - `404 model_not_found`: the id after `openai/` is not on the Models page at this base URL. Copy it from the page rather than typing it. - `404` with no model code, or LiteLLM's `NotFoundError`: `api_base` is missing `/v1`, or carries an extra path. Set it to exactly `https://api.uplinkswitch.com/v1`. - `BadRequestError: LLM Provider NOT provided`: the `openai/` prefix is missing from `model`. LiteLLM uses it to pick the OpenAI-compatible path. - `429 rate_limited`: the key hit its requests-per-minute or concurrency limit. The response carries `retry-after` in seconds. Wait that long and resend. ## Good to know - Several `model_list` entries may share one `model_name`; LiteLLM then load-balances across them. One entry per id is enough here, since one base URL already covers every model. - `litellm.embedding` takes the same `openai/`, `api_base` and `api_key` for embedding ids on the list. - Streaming is `stream=True` on `completion` or the client; nothing changes on this side. - A key per tool keeps the spend apart on the Usage page. The [LangChain](https://uplinkswitch.ai/guides/langchain), [OpenCode](https://uplinkswitch.ai/guides/opencode) and [Claude Code](https://uplinkswitch.ai/guides/claude-code) guides give each tool its own settings on the same balance. - LiteLLM's docs for the compatible provider are at [docs.litellm.ai](https://docs.litellm.ai/docs/providers/openai_compatible); the config on this page was read from them on 2026-10-28. ## Questions ### Yes. Put openai/ in front of the model id and pass api\_base and api\_key to litellm.completion, or the same three fields under litellm\_params in the proxy config.yaml (LiteLLM docs, read 2026-10-28). With api\_base set to https://api.uplinkswitch.com/v1, every chat model on the Models page answers, Claude and Gemini included. ### https://api.uplinkswitch.com/v1, ending at /v1. LiteLLM's docs say a Not Found error on a compatible endpoint usually means the /v1 postfix is missing, and that no further path belongs on api\_base because the client appends /chat/completions itself. ### Yes. The Claude ids answer the OpenAI chat completions format here, so model="openai/claude-sonnet-5-5" with this api\_base is the whole setup. No anthropic/ prefix and no ANTHROPIC\_API\_KEY are involved. ### Set api\_key to os.environ/UPLINKSWITCH\_API\_KEY under litellm\_params and export the variable before you start the proxy. LiteLLM reads the variable from the environment (docs, read 2026-10-28); the file carries only the variable name. ### Whatever the model page says per million tokens, live from the catalog; the table above has today's input and output rate for 6 common picks. Claude Sonnet 5.5 is $1.20 in and $3.00 out right now. Each request shows on the console Usage page within seconds with its tokens and charge, and a non-streamed response carries x-uplink-charge in its headers. ### 401 invalid\_api\_key means the key is wrong or revoked: copy it again from the console. 402 insufficient\_balance means the prepaid balance does not cover the request: top up in the console. 402 monthly\_limit\_reached means the key hit the monthly spend limit you set on it. Get a key in the [console](https://uplinkswitch.ai/console): prepaid credit, no subscription, 30-day price lock. ## Related - [SDK setup](https://uplinkswitch.ai/guides/sdks) - [LangChain setup](https://uplinkswitch.ai/guides/langchain) - [Vercel AI SDK setup](https://uplinkswitch.ai/guides/vercel-ai-sdk) - [OpenHands setup](https://uplinkswitch.ai/guides/openhands) - [OpenRouter vs LiteLLM](https://uplinkswitch.ai/compare/openrouter-vs-litellm) - [Claude Sonnet 5.5](https://uplinkswitch.ai/models/claude-sonnet-5-5) - [GPT-6.1 Sol](https://uplinkswitch.ai/models/gpt-6.1-sol) - [DeepSeek V4 Pro](https://uplinkswitch.ai/models/deepseek-v4-pro-0813) - [Model catalog](https://uplinkswitch.ai/models) - [LLM API price index](https://uplinkswitch.ai/pricing-index)