THIS WEEK IN AI: Qwen 3.6 is Scary Good, OpenAI's Record-Shattering Fundraise, Oracle Layoffs

This week's Limitless Podcast covers rapid AI developments, with a strong focus on China's advancing models rivaling US frontiers, OpenAI's unprecedented fundraising, and updates from Anthropic, Google, and others. Hosts discuss benchmarks, demos, funding, and efficiency moves amid a fast-moving field.

Chinese Models Signal Catch-Up

Alibaba's Qwen released Qwen 3.6 Plus and 3.5 Omni, both closed-source—a shift from prior open-sourcing that hosts see as evidence China no longer needs to commoditize US leads. Qwen 3.6 matches or beats Claude 4 Opus on some benchmarks like Terminal Bench 2.0 for coding (while trailing on broader software engineering), runs 2-3x faster, costs less (free for now), and has a 1M token context window. Qwen 3.5 Omni handles multimodal inputs: a demo turns a hand-sketched website video into functional code via "audio visual vibe coding"; another shows real-time interruption handling (ignores "mm-hmm" affirmations while summarizing a scrolled research paper via phone camera). Zhipu AI's GLM model independently demos similar sketch-to-code, suggesting coordinated progress across labs. Hosts note developer retention issues (e.g., Inflection/Meta talent restricted) but emphasize these as signs China has caught up, reducing reliance on US designs.