Qwen3.8-Max is Alibaba's flagship large language model and the largest open-weight release to date, built for developers and enterprises that want frontier-level multimodal reasoning without per-seat vendor lock-in. Tongyi Lab shipped it on August 3, 2026, with the full 2.4-trillion-parameter weights (about 95B active per token) scheduled to open-source the following week. It runs a 1M-token context window and ingests text, image, video, and audio in one pass.
Core Features
- 2.4T-parameter mixture-of-experts architecture with roughly 95B active parameters per token for strong throughput on consumer-grade clusters.
- Native multimodal context: text, image, video, and audio share a single 1M-token window instead of separate models.
- Long-horizon agentic coding — Alibaba reports the model coded autonomously on a real software project for 16 days during internal testing.
- Benchmark standing of #6 across 218 models on the benchlm.ai composite at launch (79.8/100), ahead of several closed rivals.
- Open-weight license so teams can self-host, fine-tune, and serve privately rather than call a closed API.
Use Cases
- Engineering teams running long agentic coding sessions that need a 1M-token window for whole repos.
- Enterprises that must keep weights on their own hardware for data-sovereignty or cost control.
- Multimodal product search and document understanding that mixes PDFs, images, and video.
- Research groups benchmarking open vs closed models on reasoning and agent tasks.
Pricing
Qwen3.8-Max is available through Alibaba's API at $2 per million input tokens and $6 per million output tokens, with open weights free to download and self-host. That positions it as one of the cheaper frontier-class models once you run volume.
Our Take
Best for teams that want Claude-Opus-class breadth but refuse to rent it forever — the trade-off is operational: you own the weights, so you also own the GPUs, inference stack, and uptime. For most, the hosted API is the simpler on-ramp. Browse more foundation models in our AI Engine/Model directory.
