← Gallery / Trend 05 — Dark Mesh Fusion FUSION 02 × 03
refs: anthropic.com · runway.ml · elevenlabs.io · cursor.com
aurora-3.1-fast · ships today · read paper →

Inference, at the edge of light.

Run frontier models 4× faster than the open APIs. Stream tokens in 80ms. Pay only for what you spend, never what we provision.

Get started — free $50 credit $ curl aurora.dev/install.sh | sh
TRUSTED BY
LINEAR vercel FIGMA notion RAYCAST
PROTOCOL

Five lines of code. Production-ready.

OpenAI-compatible API. Drop-in replacement. Streaming, function calls, vision — all the things you'd expect, none of the rate limits.

stream.ts · aurora-sdk — 14:23
streaming
// 1. Import the SDK
import { Aurora } from 'aurora-sdk' ;
// 2. One client, every model
const ai = new Aurora ({ apiKey: process.env.AURORA_KEY });
// 3. Stream tokens — same shape as OpenAI
const stream = await ai.chat.stream({
model: 'aurora-3.1-fast',
messages: [{ role: 'user', content: 'Write a haiku about caching.' }]
});
// 4. Token by token, ~80ms first-byte
for await ( const chunk of stream) {
process.stdout.write(chunk.text);
}
→ output stream
Memory keeps the page
Quietly between two thoughts —
Cache hit, breathing slow.
82ms ttft 847 tok/s $0.00012 total
AT SCALE · LAST 30 DAYS

The numbers we're proud of.

requests
2.4B
served · 99.999% SLA
latency
82ms
median ttft · global
savings
74%
cheaper vs OpenAI direct
regions
14
data centers · 6 continents
PLATFORM

Everything you need. Nothing you don't.

01 / MODEL CATALOG

Every frontier model. One endpoint.

FAST
aurora-3.1
847 tok/s
SMART
aurora-3.1-pro
312 tok/s
VISION
aurora-3.1-v
multimodal
EMBED
aurora-embed-2
1536 dim
CODE
aurora-code
256k ctx
+24 more →
02 / PRICING
from $0.20 / 1M tokens
74% under open API price
03 / COMPATIBLE

Drop-in OpenAI replacement.

Change base_url. Ship.
04 / CACHING

Automatic prompt caching

90% cheaper · 4× faster on hits
05 / OBSERVABILITY

Per-request traces

06 / GET STARTED
Free $50 credit.
No card. 30 days.
FROM A USER, UNPROMPTED
"We migrated 14B monthly tokens to Aurora in a single afternoon. Bill dropped 71%. Latency fell to a third. I keep waiting for the catch."
Eleni Vassilakis
CTO · Mantle · ex-OpenAI

Stop renting GPUs. Start shipping intelligence.

Most teams move their first request through Aurora in under 4 minutes. Some do it in 90 seconds.

Get API key — free $50 credit Read the docs first →