Qwen 3.8 Max: 2.4T Parameter Multimodal Model Overview

Model Architecture and Performance

Qwen 3.8 Max is a 2.4 trillion parameter multimodal model with 95 billion active parameters. It is currently positioned as a top-five contender on the LLM Arena, performing comparably to Claude 3.5 Opus in coding and agentic benchmarks. While the Max version is intended for API usage, there is significant developer interest in an upcoming 27B parameter variant designed to run on consumer-grade hardware.

Agentic Capabilities and Long-Horizon Tasks

The model demonstrates advanced agentic behavior, specifically in long-horizon task execution. It can generate and evolve its own execution harnesses, such as the open-source "Oh My CLI" tool. In testing, the model successfully performed multi-step tasks including:

• Web App Generation: Created an ISS tracker that updates location data every 5 seconds using interleaved tool calls and Python execution.
• UI/UX Design: Generated a functional Pokémon encyclopedia with interactive animations.
• Complex Spatial Reasoning: Attempted a crowd animation task where it autonomously chose to overlay text as a shortcut to organize individual entities, demonstrating a unique form of "cheating" to satisfy prompt constraints.
• 3D Mapping: Implemented a 3D tourist map of Los Angeles with toggleable night mode and location-based navigation.

API Economics and Licensing

Qwen 3.8 Max is priced at $2 per million input tokens and $6 per million output tokens, positioning it as a cost-effective alternative to Kimi K3. While the model is expected to be released with open weights, the specific commercial licensing terms remain unconfirmed. The release highlights a growing industry trend of massive models that are primarily accessible via first-party APIs rather than local deployment.