UplinkSwitch

    • Flagship AI models.
    Up to 95% cheaper than official list price.

    Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5, GPT-6.1 Sol and 77 more models. One key, one URL, the SDK you already use.

    Get a key Try it now

    https://api.uplinkswitch.com/v1 · 81 models and prices · price your last bill

    "Keep"KeepyouryourSDK.SDK.ChangeChangethetheURLURLandandthethekey.key.PayPaylesslessforforeveryeverymodel,model,andandseeseeeveryeverycall."call."

    Try it.

    $1.20 in · $3.00 out per M$10.00 −70%
    $0.60 in · $3.00 out per M$10.00 −70%

    Claude, GPT, Gemini, Grok, DeepSeek, Qwen and more on one key. The OpenAI, Anthropic and Gemini SDKs and Claude Code: change the URL and the key.

    Claude
    OpenAI
    Gemini

    Built for teams.

    Low latency from anywhere. Requests enter at the nearest Cloudflare location.

    Every model in one list. 81 models from Anthropic, OpenAI, Google, xAI, DeepSeek and 9 more labs.

    "Your key is fine, but the model isn't responding right now. Nothing was charged."

    Plain errors

    A request that never reached the model costs nothing.

    See what every call cost. Model, tokens and charge for every request. Example figures.

    Nothing hidden in the response. Every reply names the request, the model and the charge.

    Pay less than list. 81 models, most of them 40% to 70% under the developer's price. See them all. Top-ups from $10.

    95%

    Up to, below official list price on output

    40%

    At least, below the list price on output

    30 days

    Price lock on every top-up

    $10

    Minimum top-up

    Chart: sample data.

    Open source. An AI SDK provider on npm and examples for every tool, at github.com/uplinkswitch.

    terminal
    $ npm install ai @uplinkswitch/ai-sdk-provider
    $ export UPLINKSWITCH_API_KEY="sk-uplink-live-..."
    import { createUplinkSwitch } from "@uplinkswitch/ai-sdk-provider";
import { generateText } from "ai";

const uplinkswitch = createUplinkSwitch(); // reads UPLINKSWITCH_API_KEY

const { text } = await generateText({
  model: uplinkswitch("gpt-6.1-sol"),
  prompt: "Say hello in five words.",
});

    Four routes, one rate card. Each model through the SDK it was built for.

    Anthropic models · managed route

    Claude

    Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5. 12 models. Up to 70% under list. Anthropic SDK and Claude Code.

    OpenAI models · managed route

    GPT

    GPT-6.1 Sol, GPT-6 Sol, GPT-5.5. 13 models. Up to 70% under list. OpenAI SDK.

    Google models · managed route

    Gemini

    Gemini 4 Argon, Gemini 3.8 Flash, Gemini 3.5 Flash. 12 models. Up to 70% under list. Gemini SDK.

    Other labs · managed route

    Other labs

    Grok 4.6, DeepSeek V4 Pro 0813, Kimi K3, Qwen3.7 Max and 40 more from xAI, DeepSeek, Moonshot AI, Qwen and others.

    Common questions. Something else? Contact us.

    One API key and one URL for Claude, GPT, Gemini and the open-weight labs, below list price.

    No. One UplinkSwitch key calls every model, with no Anthropic, OpenAI or Google account needed.

    We buy capacity wholesale and pass most of it on. No seat fees, no minimums.

    No. We keep counts and charges, not content. Two exceptions: Replay, only for keys you turn it on for, and generated media files, kept 30 days so you can download them.

    From the provider's response, and never more prompt tokens than your request could hold. Estimated only when the provider sends none, and marked.

    No. A refused answer costs nothing, up to $5 of refunds per account per day.

    Ask with a JSON schema, JSON mode or the header x-uplink-require: json. If a non-streamed answer does not match, we retry once and do not bill the miss, within the same $5 of refunds per day.

    The OpenAI, Anthropic and Gemini SDKs, Claude Code, Cursor, and any HTTP client.

    Prepaid, from $10, in the console. Each top-up locks your prices for 30 days.

    Everything on the Models page. It is generated from the live catalog.

    Yes, for committed monthly volume. Tell us what you expect to run.

    Yes. Each key gets its own limits, models and spend cap, and stops within a second when revoked.

    If a model stops answering, UplinkSwitch retries the request automatically, at the same price. A request that fails is not charged.

    Get your key.

    Sign in, load a balance, change two settings.

    Open the console