QWEN · SERVERLESS AND DEDICATED · LIVE
qwen3-235b-a22b-2507
The 2507 refresh of Qwen's 235B mixture of experts, 22B active per token, served streamed byte for byte.
262K
TOKEN CONTEXT
$0.0675
INPUT, PER MILLION
$0.4125
OUTPUT, PER MILLION
Streamed
BYTE FOR BYTE, NO BUFFER
Overview
qwen3-235b-a22b-2507 is the July 2025 refresh of Qwen's largest text model, a sparse mixture of experts with 235 billion total parameters and 22 billion active per token. The 2507 revision improved instruction following, math, and long-context behavior over the spring release, and it remains one of the strongest open-weight reasoners in the catalog.
The economics change how teams use it. At $0.0675 per million input tokens it is cheaper than the median listing for a frontier-class open model, cheap enough to leave reasoning on for every request: agents that think on each step, analysis pipelines that never downgrade to a small model, multilingual traffic at production volume. The 262,144-token window holds a full working set without summarizing.
Serving on Nyx
Requests stream from the shared serverless endpoint the moment tokens exist, byte for byte with no chunk buffering in between. When declared capacity is exhausted the API answers 429 immediately with a Retry-After header rather than holding your connection open into a timeout.
Prefix caching carries context across turns and removes 31 percent of prefill work on agent traces, which matters for a model that invites 262K-token prompts. Dedicated pods pin capacity at a flat hourly price when you need 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": "qwen3-235b-a22b-2507",
"stream": true,
"messages": [{"role": "user",
"content": "Work through this rate-limit math and check each step."}]
}'