Alibaba Releases Qwen3.6 Max Preview: 1 Trillion Parameter MoE Model With 262K Context Window
Alibaba Cloud has released Qwen3.6 Max Preview, a proprietary frontier model built on sparse mixture-of-experts architecture with approximately 1 trillion total parameters. The model supports a 262,144-token context window and features integrated thinking mode for multi-turn reasoning, priced at $1.30 per million input tokens and $7.80 per million output tokens.
Qwen3.6 Max Preview — Quick Specs
Alibaba Releases Qwen3.6 Max Preview: 1 Trillion Parameter MoE Model With 262K Context Window
Alibaba Cloud has released Qwen3.6 Max Preview, a proprietary frontier model with approximately 1 trillion total parameters using a sparse mixture-of-experts (MoE) architecture. The model supports a 262,144-token context window and is priced at $1.30 per million input tokens and $7.80 per million output tokens.
Technical Specifications
According to Alibaba, Qwen3.6 Max Preview is optimized for agentic coding, tool use, and long-context reasoning. The model includes an integrated thinking mode that preserves reasoning traces across multi-turn conversations, similar to approaches seen in other reasoning-capable models.
The model supports structured output and function calling capabilities. The sparse MoE architecture activates only a subset of the 1 trillion total parameters for each inference, which typically provides efficiency advantages over dense models of equivalent size.
Availability and Access
Qwen3.6 Max Preview is available exclusively through the Alibaba Cloud Model Studio and Qwen Studio APIs. No open weights are provided. The model is also accessible through OpenRouter, which routes requests across multiple providers.
The model was released on April 27, 2025, according to the listing on OpenRouter.
Pricing Structure
- Input tokens: $1.30 per million tokens
- Output tokens: $7.80 per million tokens
The 6:1 ratio between output and input pricing is consistent with cost structures for compute-intensive inference in large language models.
Reasoning Capabilities
The integrated thinking mode allows the model to show step-by-step reasoning processes. According to OpenRouter's documentation, developers can access reasoning traces through a reasoning_details array in API responses. The model can preserve these reasoning traces when passed back in subsequent conversation turns, enabling continuous reasoning across multi-turn interactions.
What This Means
Qwen3.6 Max Preview represents Alibaba's entry into the trillion-parameter model tier, joining other frontier models in supporting extended context windows beyond 200K tokens. The sparse MoE architecture suggests a focus on inference efficiency, though Alibaba has not disclosed the number of active parameters per forward pass. The proprietary, API-only release contrasts with Alibaba's previous pattern of releasing open-weight Qwen models, indicating a strategic shift toward commercial model offerings. The emphasis on agentic coding and tool use positions this model for enterprise workflows requiring autonomous task execution.
Related Articles
KRAFTON releases Raon-1 mini preview, a 28.8B-parameter MoE model with 4.3B active and 262,144-token context
KRAFTON AI has released Raon-1 mini preview, a Mixture-of-Experts language model trained from scratch with 28.8B total and 4.3B active parameters. It is Apache 2.0 licensed, supports 262,144 tokens of context, and ships with a speculative decoding draft model that KRAFTON says reaches up to 2.18× higher output throughput on a single NVIDIA B300.
DeepSeek V4.1 Flash 1017: $0.15/$0.6 per 1M tokens, 1M context, 67% hallucination rate on non-answers
DeepSeek has upgraded its open-weight V4.1 Flash to revision 1017, a 552B-parameter MoE model with 16B active parameters and a 1 million token context window. Off-peak API pricing is $0.15 per 1M input tokens and $0.6 per 1M output tokens. Artificial Analysis data shows strong intelligence scores but weak results on hallucination.
MiniMax releases M3.1-Flash: 428B-parameter multimodal MoE with 1M-token context and weights on Hugging Face
MiniMax has published MiniMax-M3.1-Flash on Hugging Face: a native multimodal mixture-of-experts model with about 428B total parameters, 23B activated per token, and a context window of up to 1M tokens. It builds on MiniMax-M3 with sparse attention, quantization-aware training and DSpark speculative decoding.
NVIDIA releases Nemotron 3.5 Bolt E4B 4GB, an open small model targeting 4GB of VRAM
NVIDIA released Nemotron 3.5 Bolt E4B 4GB, an open small language model post-trained from Gemma 4 E4B. It targets local voice assistants, game characters and robots, with a footprint of about 4GB of VRAM on RTX and Jetson hardware.
Comments
Loading...