Ramp enters AI inference with its new model router

Corporate card unicorn Ramp is moving fast into the AI plumbing business.

The $44 billion spend-management startup just launched Router, an API service that lets companies cycle between major frontier models from OpenAI, Anthropic, DeepSeek, xAI, Moonshot, Minimax, and Nvidia through a single interface.

Internal tooling turned into an API gateway

Ramp claims it has been dogfooding this system internally for three years to wrangle its own AI costs. Now, it is packaging that workflow for everyone else.

The feature set looks built for CFOs as much as developers. Users can route prompts based on provider flex-tier discounts, or configure automatic model selection based on up to three specific benchmarks. You can send lightweight queries to cheap models and reserve heavy compute for high-end reasoning, all while tracking latency, token burn, and fallback attempts inside a dedicated dashboard.

It is available in the U.S. and free through the end of 2026, though users still cover direct inference costs. Ramp is also handing out a $26 credit to get started.

Routing by benchmark and a default retention catch

While Router sounds a lot like OpenRouter, there is an important caveat inside the fine print: a default opt-out data retention policy.

Ramp logs model inputs, outputs, and tool calls for up to one year to improve its product. The company insists it scrubs personally identifiable information first, and users can opt out, but enterprises handling sensitive proprietary data will need to toggle that setting immediately.

Following its $750 million raise in June, Ramp clearly views this launch as a top-of-funnel play. Managing AI inference fits squarely alongside its existing token-spend monitoring tools.

Why it matters

We think this is a clever expansion playbook. Ramp understands enterprise financial anxieties better than most developer-tooling startups, and CFOs are currently panicked over opaque AI inference bills. Tying model switching directly into corporate spend controls solves a real operational headache.

For engineering teams, it is easily worth testing right now to grab the $26 credit and bypass platform fees through 2026. If you are already managing latency fallbacks manually, the benchmark-driven routing looks compelling. Just remember to check your data retention toggles before routing any confidential workloads through it.

Bigger picture, this shows fintech giants are racing to monetize AI middleware before someone else does, mirroring similar moves from Stripe. By positioning Router as a free entry point, Ramp creates a pipeline to cross-sell its core corporate cards and treasury software. The main risk is trust: developer loyalty often favors neutral hubs like OpenRouter, and a corporate spend platform training on user prompts might make technical teams think twice.