Ramp Launches a Model Router

Aug 20, 2026 02:26 AM - 1 day ago 2

router.com saves you clip and money

One endpoint, 1 bill, each exemplary — trim your AI costs by 40% connected average. The missing portion to maximize ROI.

01Every exemplary down 1 key

02Cut conclusion costs by 40%

03Scale to Trillions of tokens

Free routing done 2026 | $26 successful exemplary credits

Router was built to trim conclusion costs by matching each petition to the lowest-cost exemplary that meets your capacity needs.

Copy for agent

curl -fsSL https://agents.ramp.com/install.sh | sh && ~/.local/bin/ramp router configure

A Router CLI convention pinch Switchyard enabled, comparing a $45.62 Router tally against $297.85 for a generic frontier model

Built for CTOs.
Loved by CFOs.

Engineering gets the champion exemplary for each workload.
Finance gets little conclusion spend.

$200$150$100$50$0

1/11/91/171/252/2

40% Default60% Flex

What teams opportunity astir Router

Cost

Models (27)
  • Claude Fable 5
  • Claude Opus 5
  • Kimi K3
  • Claude Opus 4.7
  • GPT-5.5
  • GPT-5.6 Sol
  • GLM 5.2
  • Grok 4.5
  • Kimi K2.7 Code
  • Claude Opus 4.6
  • DeepSeek V4 Flash
  • Claude Opus 4.8
  • GPT-5.6 Terra
  • Claude Sonnet 5
  • GPT-5.4
  • Gemini 3.1 Pro
  • GPT-5.6 Luna
  • Claude Sonnet 4.6
  • Kimi K2.6
  • GLM 5.1
  • Qwen3.6 Plus
  • DeepSeek V4 Pro
  • Qwen3.7 Plus
  • GPT-5.4 Mini
  • Claude Haiku 4.5
  • GPT-5.4 Nano
  • GPT-4.1

Average

A benchmark built from existent work.

We built Ramp SWE-Bench from existent accumulation engineering activity because nationalist leaderboards couldn’t reply the questions we had. It gives america a clearer position of what each exemplary tin lick and astatine what cost.

Put to activity astatine Ramp.

Ramp gives Router a real-world proving ground. The lessons we study successful accumulation provender straight backmost into the product.

Production value

2.75T+

Tokens routed monthly

How Ramp trim AI costs by 25%+ connected soul workloads

Router responds to unrecorded latency and nonaccomplishment rates, cutting Ramp’s AI costs by much than 25% without sacrificing performance.

Read the blog post

NVIDIA NeMo Switchyard’s Stage Router for Coding Agents

Switchyard’s intelligent exemplary action reduces costs by 59% and tally clip by 35% without sacrificing performance.

Watch the video

“At Ramp, Router trim our wide LLM costs by 30% while making our features smarter and faster.”

Rahul Sengottuvelu

CTO, Ramp

More from the Lab

Jul 1, 2026

PorTAL: Portable Task Adaptation for LoRA

Learn a task adjustment erstwhile successful a base-agnostic form, past larboard it to caller stiff models by refitting only a bladed per-base alignment — recovering ~98% of per-task LoRA's assistance connected an unseen exemplary wrong the aforesaid family and ~94% crossed families.

Apr 21, 2026

Coding agents disregard their ain budgets

Agents can't beryllium trusted to negociate their ain token budgets. Spend power has to unrecorded successful a separate, evidence-grounded strategy extracurricular the supplier doing the spending.

Aug 27, 2025

How we built Agent Fill

The communicative down Agent Fill - an AI supplier that automatically fills retired forms by knowing context, extracting data, and navigating analyzable workflows.

FAQ

One endpoint for accessing aggregate AI models. Instead of wiring your app to 1 supplier astatine a time, you nonstop requests done our router which tin take the correct exemplary for the occupation based connected quality, cost, and availability. No lock-in. One statement to switch.

We did. It was either this aliases walk the adjacent twelvemonth pronunciation retired a longer URL connected podcasts. More importantly, we build devices that thief companies make amended spending decisions and extremity overpaying for things. AI tokens are the fastest-growing walk category, and we want each token you usage to beryllium worthy it.

Your petition goes to Ramp Router first. We’ll authenticate the petition and thief you way the usage, model, provider, and cost. We’ll way eligible requests to a much cost-efficient tier erstwhile it won’t impact quality. See our Router Strategies to prevention moreover more.

Router supports the latest models from OpenAI, Anthropic, and prime open-source models, including Kimi. We regularly adhd support for caller models arsenic they go available. See the afloat database of supported models successful our docs.

Ramp Router is free done 2026. You’ll salary database value for the tokens you use. To people free routing done 2026, your first $26 successful credits are connected us. Subject to connection terms.

There are, and they helped beryllium group want 1 endpoint for each model. What’s different here: Ramp has spent the past 3 years moving and improving this exertion connected our ain accumulation workloads, cutting our AI costs by 30%. Saving businesses clip and money is what Ramp does, truthful we built Router to springiness developers the devices Ramp utilized to little its ain AI costs. With Router Strategies, developers tin specify costs and capacity priorities for different types of requests, aliases commencement pinch Ramp’s benchmarked defaults. Router is free done 2026. You’ll salary database value for the tokens you use, and your first $26 successful credits are connected us. Subject to connection terms.

Going nonstop ties your exertion to 1 provider’s models, pricing, and merchandise cycle. With Router, you link erstwhile and usage eligible models done a azygous endpoint. Use Router Strategies to group really Router balances costs and capacity for different types of requests, aliases commencement pinch Ramp’s benchmarked defaults.

Router is for individual developers and teams successful the U.S. (with much countries coming soon) who want to get much from AI without overpaying for it. Whether you’re testing an thought connected your ain aliases building for a team, Router gives you 1 spot to activity crossed supported models. Enterprise features are coming soon.

No. You don’t request a Ramp card, a institution account, aliases moreover an LLC. Router is free done 2026. To people the occasion, your first $26 successful credits are connected us. Enter your email astatine router.com and we’ll nonstop you a sign-in link. Getting started takes 2 lines of code.

No. Ramp Router has an OpenAI and Anthropic compatible API, truthful if you’re already utilizing the OpenAI aliases Anthropic SDKs aliases different compatible model (which is fundamentally each of them!), switching should beryllium a one-line change: update your guidelines URL to Ramp Router’s endpoint.

Please spot the Ramp Router Privacy Policy for accusation connected really Ramp manages individual information. Users tin take to usage U.S.-hosted models that supply zero information retention (ZDR). Router itself stores exemplary inputs, outputs, and metadata and uses this information to amended the service, but users whitethorn move this disconnected successful settings. Some frontier models person provider-specific information retention policies — spot our position for details.

Yes. Router supports bring-your-own API keys (BYOK) for prime exemplary providers. See our method archiving for much details.

Ramp Router uses models hosted connected U.S.-based infrastructure by their underlying exemplary providers. See our method archiving for much details.

If a supplier goes down aliases rate-limits you, Ramp Router tin way eligible requests to different disposable model, truthful your app has a fallback erstwhile 1 supplier cannot service it.

Yes. Give 2 models the aforesaid punctual and comparison their responses, latency, and value broadside by side. It’s a speedy measurement to trial models earlier integrating 1 into your product.

Tokens are money.
Save both.

More