Groq
Groq hosts open-weight models on custom inference hardware. Latency is measured in tens of milliseconds for most workloads. Supercompat talks to Groq via the official
groq-sdk.Install
npm install supercompat openai groq-sdk
Minimal setup
import Groq from 'groq-sdk'
import {
supercompat,
groqClientAdapter,
completionsRunAdapter,
memoryStorageAdapter,
} from 'supercompat/openai'
const client = supercompat({
clientAdapter: groqClientAdapter({
groq: new Groq({ apiKey: process.env.GROQ_API_KEY }),
}),
storageAdapter: memoryStorageAdapter(),
runAdapter: completionsRunAdapter(),
})
const response = await client.responses.create({
model: 'llama-3.3-70b-versatile',
input: 'Give me a haiku about latency.',
})
Streaming
Groq's throughput makes streaming feel instantaneous:
const stream = await client.responses.create({
model: 'llama-3.3-70b-versatile',
input: 'Count from one to ten.',
stream: true,
})
for await (const event of stream) {
if (event.type === 'response.output_text.delta') {
process.stdout.write(event.delta)
}
}
Models
Pass any Groq-hosted model id — the full catalog is at console.groq.com/docs/models.
Some current examples:
llama-3.3-70b-versatilellama-3.1-8b-instantqwen-2.5-32bopenai/gpt-oss-120b