Hopscotch AI

Hopscotch AI: one API, every model, no markup

Hopscotch is an LLM gateway: a single OpenAI-compatible API that fronts more than 500 models from Anthropic, OpenAI, Google, DeepSeek, Moonshot, Qwen and others. You point your existing OpenAI SDK at its base URL, drop in one key, and every model in the catalog becomes a one-line model-name change away. The pitch that separates it from a crowded gateway market is pricing: buy credit from USD 5 and pay each provider’s list price per token – no markup, no fees, no subscription.

Spend limits that refuse, not alert

The cost-control story is stronger than most. Every API key gets a credit limit that resets daily, weekly or monthly, with monthly caps per teammate and workspace. A request that would cross a limit is refused before it reaches the provider – exactly what you want standing between an agent loop and your balance. The Activity log lists every request with model, serving provider, outcome, tokens, cost and latency, and exports to CSV; a Usage view breaks spend down by model, provider, key and teammate.

Named models and honest failover

Hopscotch commits to running the model you name – no silent swaps to cheaper models – and says which provider actually served each request. If a provider 429s or goes down, the request moves to another key or to the next model in your own routing profile, and your code still sees one request, one response. A playground runs one prompt on up to three models side by side, billed through your key, so you can compare answers and cost before changing code. These are vendor-described behaviors; we did not send requests through the gateway.

The catch: an opaque operator

For a service that proxies your prompts, the company behind it is unusually quiet. The site announces a USD 7.5m raise but publishes no legal entity, founders, headquarters or founding year, and we found no terms of service or privacy policy pages. There is a genuine free option – bring your own provider keys and the gateway layer costs nothing with no monthly cap – but teams with compliance requirements will need those legal gaps closed first.

Ashley Richmond’s editorial view

The pricing model is the most honest I have seen in this category, and the pre-request spend limits solve a real fear for anyone running agents. Start with the free BYOK mode and a tight key limit, and watch the Activity log. The provisional desk-research score is 4.1/5.

Desk-research disclosure: Official pages reviewed September 30, 2026. No account, API key, credit purchase or gateway requests. Company identity fields remain blank; no terms or privacy pages found. All prices and the 500+ model count are vendor-published.