OPENAI · SERVERLESS AND DEDICATED · LIVE
gpt-oss-20b
OpenAI's small open-weight mixture of experts, built for latency-critical chat and high-QPS API features, streamed byte for byte.
131K
TOKEN CONTEXT
$0.0225
INPUT, PER MILLION
$0.0975
OUTPUT, PER MILLION
Streamed
BYTE FOR BYTE, NO BUFFER
Overview
gpt-oss-20b is the smaller of OpenAI's two open-weight models: a mixture of experts with 21B total parameters and 3.6B active per token, released under Apache 2.0. It shares the 120B's schema and behavior profile, so a prompt tuned on one moves to the other without rework. The fastest streaming rate in this corner of the catalog comes from how little compute each token touches.
The fit is work where the clock matters more than the last point of quality: latency-critical chat, classification and extraction at high QPS, and draft generation in cascade setups where a larger model reviews the output. Its output price sits below the median listing for small open models.
Serving on Nyx
A fast model deserves a fast first byte. Output leaves the server as the engine decodes it, with no chunk buffering in between. Past declared capacity the API returns an immediate 429 with a Retry-After header, so your retry logic gets a signal instead of a stalled connection.
Prefix caching carries context across turns and removes 31 percent of prefill work on agent traces. For pinned capacity, dedicated pods run at a flat hourly price with guaranteed throughput.
OUTPUT PRICE · $/M · LOWER IS BETTER
Independent harness · run weekly · streamed
# OpenAI-compatible: swap the base URL
curl https://api.nyxprovider.com/v1/chat/completions \
-H "Authorization: Bearer $NYX_API_KEY" \
-d '{
"model": "gpt-oss-20b",
"stream": true,
"messages": [{"role": "user",
"content": "Classify this ticket: refund, bug, or billing?"}]
}'