We open sourced How to Deploy Your Model (HTDYM), our internal performance modeling tool!
Infrastructure for abundant intelligence

The most efficient inference for
long-horizon agents

Sail is the most cost-efficient API for the best open-source models.

Unbeatable pay-per-token pricing, with $5 in monthly free credits.

GLM-5.2DeepSeek V4 FlashKimi-K2.6gpt-oss-120bGemma 4 31B ITGemma 4 31B IT (NVFP4)Qwen3.6 35B A3BNemotron 3 Super 120B A12B BF16GLM-5.2DeepSeek V4 FlashKimi-K2.6gpt-oss-120bGemma 4 31B ITGemma 4 31B IT (NVFP4)Qwen3.6 35B A3BNemotron 3 Super 120B A12B BF16
Get started freeSee all models & pricing
Pay ~30% less for tokens
by default, using our drop-in API
(Same model vs. traditional inference provider)
FasterCost-efficient
defaultbalancedflex
Input
$0.80
Cached
$0.16
Output
$3.00
GLM-5.2 FP8
USD / 1M tokens
quickstart.py▶ Run
from openai import OpenAI

client = OpenAI(
    base_url="https://api.sailresearch.com/v1",
    api_key="YOUR_SAIL_API_KEY",
)

response = client.responses.create(
    model="zai-org/GLM-5.2-FP8",
    input="Plan and run a deep research report.",
)
Introducing

The Completion Window

Completion windows can be specified per request, allowing you to indicate your latency tolerance and drastically cut token costs for the same intelligence.

WindowSchedulingTypical use casePrice vs. other providers
default (asap)Low-latency servingInteractive UIs, human-in-the-loop~5–35% less
balancedWider scheduling windowBackground agents, subagents, and pipelines~45–65% less
flexBest-effort schedulingBatch processing, evals, offline~60–80% less
See pricingRead more

Drop-in APIs

OpenAI- and Anthropic-compatible (Responses, Chat Completions, and Messages APIs). Migrate to Sail in minutes.

Fine-tunes and RL Rollouts

Serve your own LoRA fine-tunes and run RL rollouts on Sail’s sampling path. Drop-in Tinker integration.

Usage-based, pay-per-token

Send requests immediately and pay as you go, with $5 in free credits every month.

Battle-tested

Trillions of tokens served per week, and ready to serve yours.

Scale your workloads without limits

Send us millions, billions, or trillions of tokens. No sales call required.

Sail is designed to absorb large bursts of traffic, so you can scale on your schedule.

Trusted by teams at the frontier
Parallel logo
Web search & research APIs for AI agents
Detail logo
AI-powered code review
Jack & Jill logo
Your AI career agent
Quadrillion logo
AI agent workforce for research

“We and Sail share a belief that background agents are about to do far more useful work. Getting there takes efficient, scalable inference paired with the highest-quality context, including from the web. Sail is building the inference side of that, and we’re glad to be aligned on where this is going.”

Travers Nisbet, Co-founder, Parallel

“Building on Sail lets us ship long-horizon agents with great economics. Trillions of tokens and counting — we’re happy customers.”

Dan Robinson, CEO, Detail.dev

“We’re working with Sail Research to deliver the best experience any researcher can ask for, with Sail’s infrastructure offering the most flexible compute.”

Tiffany Zhao, Head of Strategy, Quadrillion

The pursuit of token abundance.

We curate the best open-source models and invest in serving them as efficiently as physically possible, on the full latency vs. cost curve.

From CUDA to scheduling, we ensure no compute goes to waste, and no token costs more than it should.

Because more agents, with more intelligence, can do incredible things.

Our why and how
Pricing plans

Two ways to Sail.

Self-Serve
Freeto start, pay-as-you-go

Get your API key, and let the tokens flow.

Get started with $5 free credit
$5 in free credits every month when you attach a payment method
Usage-based pricing with prepaid credits
Enterprise
Custom

Workloads at extreme scale.

Talk to us
Volume pricing, billed monthly in arrears
HIPAA support with a signed BAA
Signed MSA and DPA
Region-locked data processing
Uptime and latency SLAs
Early access to new features
Dedicated support

Every agent needs a computer

Sailboxes are full VMs that give long-horizon agents compute they can run indefinitely, with wonderful economics.

Pair them with Sail inference so your agents can read, write, build, and test for hours, days, or even longer without losing state — and at maximal cost efficiency.

Explore Sailboxes

FAQ

How do I get an API key?

Sign up here and generate an API key in seconds. Point any OpenAI- or Anthropic-compatible client at our base URL and start sending requests.

How much will my agent cost to run on Sail?

It depends on your token volume, model, and chosen completion window. Use the agent cost calculator to estimate spend for your workload.

Open the agent cost calculator
Are the APIs really OpenAI- and Anthropic-compatible?

Yes. Use the official SDKs or any compatible client — just swap the base URL and key. Responses, Chat Completions, and Messages all work as expected.

What are completion windows?

Completion windows let you indicate your latency tolerance and drastically cut token costs. balanced and flex give Sail more scheduling room in exchange for lower token prices. A completion window can be optionally specified on each request to Sail's API, and the pricing page shows which windows each model supports.

Read more about completion windows
Does Sail enforce rate limits?

No strict rate limits and no limit-increase process. Sail is designed to absorb large bursts of traffic, and the flex completion window is best for maximum throughput on very large workloads.

Is Sail SOC 2 compliant?

Sail is SOC 2 Type I compliant, and SOC 2 Type II compliance takes effect August 25, 2026. You can review our security controls and request the Type I auditor report in our Trust Center.

Visit the Trust Center
How does Sail handle inference data?

We have Zero Data Retention (ZDR) by default, and use inference request and response data only to provide the service and never train models on it without your written consent. Our DPA covers the default terms, and enterprise customers can sign a DPA.

Read about security and privacy
Where is customer data processed or stored?

Sail uses service providers in multiple regions. Enterprise customers can pin traffic to a specific geographic region.

Read about security and compliance
Does Sail support HIPAA-regulated workloads?

Yes. Enterprise customers can sign a BAA for HIPAA-regulated workloads, along with an MSA and DPA.

Read about security and compliance
How do you make inference so efficient?

We work at every level of the stack:

  • Writing CUDA to push toward speed-of-light performance on GPUs
  • Digging into the guts of inference engines like SGLang to maximize efficiency
  • Distributing work across providers to maximize robustness and fleet utilization
  • Using spot compute when it's available, and safely failing over to more reliable compute when it's not
Which models are supported?

The leading open models, listed here, with support for LoRA fine-tunes.

Do you support fine-tunes? RL rollouts?

Yes. Bring your own LoRA adapters and run them on supported models. If you train LoRAs with Tinker, you can also sample directly from Tinker checkpoints without uploading, and use Sail as a drop-in TokenCompleter in Tinker.

How is pricing structured?

Usage-based, pay-per-token, and extremely competitive, with $5 in free credits refreshed every month. Enterprise contracting is also available (contact us).