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.

Provider A188
Provider B137

OUTPUT PRICE · $/M · LOWER IS BETTER

Nyx$0.4125
Provider A$0.55
Provider B$0.66

Independent harness · run weekly · streamed

Nyx$0.0675$0.4125241
Provider A$0.09$0.55188
Provider B$0.108$0.66137
# 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."}]
  }'
CONTEXT262,144 tokens
INPUT$0.0675 / M tokens
OUTPUT$0.4125 / M tokens
STREAMINGByte for byte, no buffer
PARAMETERS235B total · 22B active
MODALITYText in · text out
LICENSEApache 2.0
DEPLOYMENTServerless · Dedicated
APIOpenAI-compatible
CATALOG

All models in the library