qwen
Qwen: Qwen3 30B A3B Instruct 2507
qwen/qwen3-30b-a3b-instruct-2507
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
qwen/qwen3-30b-a3b-instruct-2507 is Qwen's updated instruction-tuned mixture-of-experts model with 30.5 billion total parameters and 3.3 billion active per token. It handles text only, runs exclusively in non-thinking mode, and natively supports a 262,144-token context, with an optional configuration extending toward 1 million tokens. Qwen reports gains in instruction following, reasoning, coding, multilingual coverage, and long-context understanding over the earlier non-thinking release; its benchmark table is vendor-reported, so independent validation on your workload is advised. Released under Apache-2.0 per the model card.[1][2]
Strengths
- Large native context window of 262,144 tokens, with a documented path to roughly 1 million tokens via dual chunk attention and sparse attention.[2]
- Sparse MoE design activates only 3.3B of 30.5B parameters per token, which Qwen positions as efficient inference.[2]
- Broad API parameter support per the OpenRouter catalog, including tool calling, structured outputs, and logprobs; the card also highlights tool-use capability.[1][2]
Limitations
- Text-only modality and non-thinking mode only; the card states it never generates think blocks.[1][2]
- Performance evidence is vendor-reported (benchmark table plus a RULER long-context test); treat claimed gains as assertions until independently validated, and the vendor's agentic-task results are uneven.[2]
- The ~1M-token setup requires replacing the config file and roughly 240 GB of GPU memory; even the default context may need memory tuning per the card's OOM guidance.[2]
Where to use it
- Potentially well suited to long-document work such as summarization or retrieval over large corpora, given the 262K native context.[2]
- A reasonable candidate for instruction-following, multilingual, or coding assistants where a non-thinking MoE checkpoint fits latency and cost constraints.[1][2]
- Tool-using agents via Qwen-Agent or OpenAI-compatible servers (vLLM, SGLang) are documented; validate agentic reliability on your own tasks.[2]
Sources
- [1] OpenRouter catalog · Qwen: Qwen3 30B A3B Instruct 2507Sources checked: Nov 13, 2026
- [2] Model documentation · Qwen: Qwen3 30B A3B Instruct 2507Sources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
Qwen3-30B-A3B-Instruct-2507 is a 30.5B-parameter mixture-of-experts language model from Qwen, with 3.3B active parameters per inference. It operates in non-thinking mode and is designed for high-quality instruction following, multilingual understanding, and...
Specifications
- Model ID
- qwen/qwen3-30b-a3b-instruct-2507
- Creator
- qwen
- Added to OpenRouter
- 29.07.2025
- Context length
- 262,144
- Modality
- text->text
- Input modalities
- text
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.100 / out $0.300
- Knowledge cutoff
- 30.06.2025
- Expiration date
- —
Tasks and capabilities
A grade appears only where the source supplies a matching measurement for that task, the model is active and the cohort holds at least 10 comparable models. The capability column lists facts the source declares — supported formats, reasoning and tool parameters — which are not measured quality.
| Task | Grade | Source-declared capability |
|---|---|---|
| General quality | No measurement | — |
| Coding | No measurement | — |
| Agents | No measurement | parameters: structured_outputs, tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | — |
| Document analysis | No measurement | — |
| Vision | No measurement | — |
| Image generation | No measurement | — |
| Audio | No measurement | — |
| Video | No measurement | — |
| Embeddings | No measurement | — |
Evaluations
No evaluations are supplied for this model.
How grades work
Each letter grade compares a model only against other models scored by the same source and metric in the current snapshot. S means the top 10%, A the next 20%, B the next 30%, C the rest. Models without a published measurement are shown as not rated — a supported format alone never implies measured quality.
- S — S — top 10% of the cohort
- A — A — next 20% of the cohort
- B — B — next 30% of the cohort
- C — C — below the B band
Providers
Optional live provider statistics from OpenRouter. Missing values are unknown, not zero.
| Provider | Context length | Price in /1M | Price out /1M | Latency p50 (30m, s) | Throughput p50 (30m, tok/s) | Uptime (30m) |
|---|---|---|---|---|---|---|
| DekaLLM | 262,144 | $0.090 | $0.300 | — | — | 100% |
| SiliconFlowfp8 | 262,144 | $0.090 | $0.300 | — | — | 100% |
| Nebiusfp8 | 262,144 | $0.100 | $0.300 | — | — | 83.33% |
Provider data retrieved: 13.11.2026
Related articles
No published articles identify this model yet.