Qwen 3.8 Max and the Geopolitics of Open-Source AI

Performance and Capabilities

Alibaba's Qwen 3.8 Max is a 2.4-trillion parameter model that demonstrates performance competitive with US-based frontier models like Fable and GPT-5.6 Soul. The model excels in coding benchmarks, specifically Terminal Bench, where it scores 86.6 compared to Fable's 84.6. Alibaba highlights the model's ability to perform autonomous research by taking a research paper, writing code to reproduce the results, and iteratively testing improvement ideas through a recursive loop. Additionally, the model has demonstrated the ability to execute the entire silicon design flow, a capability that potentially addresses China's historical weaknesses in chip manufacturing and design.

Economic and Geopolitical Implications

Qwen 3.8 Max is significantly cheaper than its US counterparts, priced at $2 per million input tokens and $6 per million output tokens, compared to Fable's $10 and $50 respectively. While these lower costs provide enterprise users with alternatives to expensive closed-source APIs, the author notes that the true cost of intelligence is determined by the number of tokens required to complete a task rather than the price per token alone.

There is a growing geopolitical tension regarding the reliance on Chinese open-source models. While US companies can host these models locally to avoid sending data to China, the long-term risk involves the co-design of models with specific hardware. If Chinese labs optimize their models for proprietary hardware that US firms eventually adopt, it could create a dependency on Chinese infrastructure. The author remains conflicted on whether these open-source releases will successfully commoditize the AI layer or if the massive compute advantage and recursive self-improvement capabilities of US labs like OpenAI and Anthropic will render the open-source competition irrelevant.