Drop-in compatible
Keep your SDK. Swap the base URL and every endpoint just works.
- baseURL: "https://api.other.com/v1"
+ baseURL: "https://api.ozmio.dev/v2"
ozmio-1-fast — now 2.1× faster at the same price — 1,840 tokens/s on edge. is 2.1× faster
See benchmarksPOST /v2/chat/completions — 200 OK · 142 ms
OpenAI-compatible inference for 60+ frontier and open models, served from 14 edge regions with automatic fallbacks. Change one line, pay per token.
Model
curl https://api.ozmio.dev/v2/chat/completions \
-H "Authorization: Bearer $OZMIO_KEY" \
-d '{
"model": "ozmio-1-fast",
"stream": true,
"messages": [{ "role": "user",
"content": "Name 3 uses for edge inference" }]
}'
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.ozmio.dev/v2", // ← the only change
apiKey: process.env.OZMIO_KEY,
});
const stream = await client.chat.completions.create({
model: "ozmio-1-fast", stream: true,
messages: [{ role: "user", content: "Name 3 uses for edge inference" }],
});
from openai import OpenAI
client = OpenAI(base_url="https://api.ozmio.dev/v2")
stream = client.chat.completions.create(
model="ozmio-1-fast",
stream=True,
messages=[{"role": "user",
"content": "Name 3 uses for edge inference"}],
)
⌘ ↵ to send
Response · stream
waiting
01 Models
Sort by price, speed or context. Every model speaks the same request format.
| Type | |||||
|---|---|---|---|---|---|
| ozmio-embed-3 | Embeddings | 8K | $0.02 | — | 9,000 t/s |
| ozmio-1-fastNew | Text | 128K | $0.10 | $0.40 | 1,840 t/s |
| open-8b-instruct | Text | 32K | $0.12 | $0.12 | 2,100 t/s |
| ozmio-coder-2 | Code | 256K | $0.30 | $1.20 | 920 t/s |
| open-70b-instruct | Text | 128K | $0.59 | $0.79 | 640 t/s |
| ozmio-vision-1 | Vision | 64K | $0.80 | $2.40 | 410 t/s |
| ozmio-1-large | Text | 200K | $2.00 | $8.00 | 310 t/s |
| open-405b-reasoning | Code | 128K | $3.00 | $3.00 | 120 t/s |
Prices in USD. Cached input tokens are billed at 25%. Batch requests are 50% off.
02 Benchmarks
Same 70B model, same prompt, measured from 6 continents every 10 minutes. Methodology and raw data are public.
See live status2.7×
faster TTFT
640
tokens / s
−58%
cost vs. A
03 Platform
Keep your SDK. Swap the base URL and every endpoint just works.
- baseURL: "https://api.other.com/v1"
+ baseURL: "https://api.ozmio.dev/v2"
Route by cost or latency and fail over across models and regions automatically.
model: ["ozmio-1-fast", "open-70b"]
route: "lowest-latency"
Long system prompts are cached at the edge. Hits cost 75% less and return faster.
x-ozmio-cache: HIT saved: 11,204 tokens
Pass a JSON schema; get valid JSON back. Guaranteed by constrained decoding.
response_format: { type: "json_schema" }
Queue millions of requests and get results within 24 hours at half the price.
POST /v2/batches → −50%
Prompts and completions are never stored or used for training. SOC 2 Type II.
retention: "none" region: "eu-only"
04 Status
All systems operational
Live p50 latency from your nearest PoP
05 Pricing
No seats, no minimums. Commit for volume discounts when you’re ready.
Estimated monthly cost
$0
0 tokens / month
quorbit
“We cut p95 latency from 2.4 s to 610 ms by switching one base URL. Our autocomplete finally feels instant.”
Lumenfield
“Fallbacks saved us during two provider outages last quarter. Customers never noticed a thing.”
Halvard
“Batch plus caching took our nightly enrichment job from $9k to $2.1k a month, same model.”
$ npm i openai
$ export OZMIO_KEY=oz_live_••••••••
$ npx ozmio chat "hello"