Router DeskHomeQwen3 235B
Qwen3 235B
The full name is Qwen3 235B A22B. Only 22 billion parameters run on each token. The slug is qwen/qwen3-235b-a22b.
Updated 2026-09-27
This is a mixture-of-experts model: 235 billion parameters stored, 22 billion active. OpenRouter’s card says it can switch into a thinking mode for math, code, and other hard prompts. Context on 27 Sep 2026 was 131,072 tokens. Price was $0.455 per million input tokens and $1.82 per million output tokens.
Later Qwen3 cards with “2507” in the id are updated checkpoints at different prices. qwen/qwen3-235b-a22b-2507 was $0.0875 / $0.35 that same day. If you want the original A22B id from the search, use the slug without 2507. If you want the cheaper later checkpoint, use the 2507 id and say so in your notes.
Compared with the dense Qwen3 32B ($0.08 / $0.28), the 235B costs several times more and is the one to try when 32B stops following a multi-step spec. Compared with Qwen3.7 Max, the context is shorter.
Questions
What does A22B mean?
About 22 billion parameters are active on each forward pass, out of 235 billion total.
How much is Qwen3 235B?
qwen/qwen3-235b-a22b was $0.455 input and $1.82 output per million tokens on 27 Sep 2026.
Is the 2507 slug the same model?
It is a later checkpoint with its own price. Do not treat the two ids as interchangeable in an eval.
Related
- Qwen3 32B
Qwen3 32B is a dense 32.8B model. On 27 Sep 2026 OpenRouter priced qwen/qwen3-32b at $0.08 input and $0.28 output per million tokens.
- Qwen3 Max
Qwen3.7 Max was $1.475 input and $4.425 output per million tokens on 27 Sep 2026, with a 1M context. The older qwen/qwen3-max slug was $0.78 / $3.90.
- Open-weight models
Open-weight models you can call on OpenRouter in Sep 2026 include DeepSeek V4, Qwen3 32B and 235B, GLM 5.2, and the free Gemma and Nemotron slugs.
- GLM 5.2 API
GLM 5.2 on OpenRouter was $0.650 input and $2.042 output per million tokens on 27 Sep 2026. The slug is z-ai/glm-5.2, with a 1M context.