Inference at the speed of thought.

One API for eight models. First streamed token in about 120 milliseconds, prices you can read without a sales call, and a working request in under three minutes.

Get API key Read the quickstart

$5 free credit every month · no card needed

bash: first request

$ curl api.synapse.example/v1/chat \

-H "Authorization: Bearer $SYNAPSE_KEY" \

-d '{"model": "syn-flash", "stream": true,

"messages": [{"role": "user",

"content": "Explain latency in 8 words"}]}'

HTTP/2 200 OK · first token 112ms

data: {"delta": "Latency:"}

data: {"delta": " the"}

data: {"delta": " pause"}

data: {"delta": " between"}

data: {"delta": " asking"}

data: {"delta": " and"}

data: {"delta": " knowing."}

data: [DONE]

$ curl api.synapse.example/v1/chat \

-H "Authorization: Bearer $SYNAPSE_KEY" \

-d '{"model": "syn-flash", "stream": true,

"messages": [{"role": "user",

"content": "Explain latency in 8 words"}]}'

HTTP/2 200 OK · first token 112ms

data: {"delta": "Latency:"}

data: {"delta": " the"}

data: {"delta": " pause"}

data: {"delta": " between"}

data: {"delta": " asking"}

data: {"delta": " and"}

data: {"delta": " knowing."}

data: [DONE]

>Latency: the pause between asking and knowing.11 tokens · 0.21s

p50 latency —uptime 90d —requests/day —

Connecting… · status page ↗

first call

Ten lines to your first streamed token.

The same request in Python, JavaScript, Go and cURL. Pick a language once and every code sample on the site follows.

from synapse import Synapse

client = Synapse()  # reads SYNAPSE_API_KEY

stream = client.chat.create(
    model="syn-flash",
    messages=[{"role": "user", "content": "Explain latency in 8 words"}],
    stream=True,
)

for chunk in stream:
    print(chunk.delta, end="", flush=True)

platform

Everything a production workload needs.

Streaming
Server-sent events from the first token. p50 time to first token is 120ms on Flash.
Function calling
Declare tools as JSON Schema. Arguments come back validated, with parallel calls supported.
Batch API
Send up to 50,000 requests in one file. Half price, results within 24 hours.
Fine-tuning
LoRA fine-tunes on Mini, Flash, Code and Pro. Upload JSONL, deploy in minutes.
Regions
Pin traffic to us-east, eu-west or ap-southeast. Data stays where you send it.
SOC 2 Type II
Audited annually. Zero data retention on request; we never train on your traffic.

models

From fractions of a cent to a million tokens of context.

All 8 models

Synapse Flash
syn-flashtextvision
Our fastest general model. The default for chat and agents.
context
128K
in / 1M
$0.10
out / 1M
$0.40
Synapse Code
syn-codetextcode
Tuned for completion, refactors and repository-scale context.
context
256K
in / 1M
$0.60
out / 1M
$2.40
Synapse Pro
syn-protextvision
Balanced reasoning and vision for production workloads.
context
200K
in / 1M
$1.20
out / 1M
$4.80
Synapse Ultra
syn-ultratextvisioncode
Our most capable model, with a one-million-token window.
context
1M
in / 1M
$4.00
out / 1M
$16.00

pricing

Know your bill before you write code.

Per-token pricing, no seats, no minimums. Move the slider to see a month.

Model

Requests per day

Estimated monthly cost

$55.00

300K requests over 30 days on Synapse Flash.

Assumes 800 input and 300 output tokens per request.

Get API key Full breakdown

teams shipping on synapse

fernhollow
QUANTA/LABS
paperplane
Orbitly
hexfield
Lumen & Co

Your API key is three minutes away.

Get API key See pricing

Inference at the speed of thought.

Ten lines to your first streamed token.

Everything a production workload needs.

Streaming

Function calling

Batch API

Fine-tuning

Regions

SOC 2 Type II

From fractions of a cent to a million tokens of context.

Synapse Flash

Synapse Code

Synapse Pro

Synapse Ultra

Know your bill before you write code.

Estimated monthly cost

Your API key is three minutes away.