Explore additional models
Frequently
asked questions
Qwen3.6 35B A3B is Alibaba’s open mixture-of-experts model, with 35B total parameters and about 3B active per token, so it runs at near small-model latency while retaining mid-size quality. It supports a 262,144-token context and ships under the Apache 2.0 license. It is available for Serverless Fine-Tuning and high-performance inference via Self-Serve Deployments in Crusoe Intelligence Foundry.
Qwen3.6 35B A3B is a mixture-of-experts model with about 3B active parameters, so it is faster and cheaper per token to serve. Qwen3.8 27B is a newer dense model that uses all 27B parameters on every token, delivers higher peak quality, and also accepts image input. Choose Qwen3.6 35B A3B for high-throughput production and Qwen3.8 27B for quality-sensitive or visual work.
Qwen3.6 35B A3B is billed by how you use it. Serverless Fine-Tuning is billed per 1M training tokens, where training tokens equal dataset tokens multiplied by epochs. Self-Serve Deployments are billed per GPU hour based on the GPU you choose, giving you predictable costs at production volume. Current rates are shown in the pricing section above.
Alibaba reports Qwen3.6 35B A3B matching or beating the earlier Qwen3 235B A22B Instruct on many benchmarks, with roughly one-seventh the active parameters and a fraction of the memory. That makes Qwen3.6 35B A3B the more efficient choice for production serving, while Qwen3 235B A22B Instruct offers a larger knowledge base for fine-tuning. Both are available for Serverless Fine-Tuning and Self-Serve Deployments on Crusoe.
Yes. Use Serverless Fine-Tuning to customize Qwen3.6 35B A3B on your own data. Upload a JSONL or Parquet dataset, run a LoRA job with pre-configured defaults, then go live in one click using Self-Serve Deployments or download the weights under Apache 2.0. Because only about 3B parameters are active per token, the fine-tuned model serves at small-model speed. Self-Serve Deployments offer inference profiles optimized for throughput or responsiveness. You can also request a Tailored Deployment for bespoke inference optimization and a dedicated endpoint.
The “A3B” in Qwen3.6 35B A3B stands for about 3 billion active parameters. The model uses a mixture-of-experts architecture with 35B total parameters, but it routes each token through only about 3B of them. That gives it mid-size knowledge while computing like a much smaller model, which keeps latency and serving costs low at high request volumes.


%204.png)
