TLDRocket
30 July 2026
The open-model insurgency is reshaping AI economics faster than incumbents can lobby against it. Moonshot's Kimi K3—a 2.8 trillion-parameter vision-language model priced at $3 per million tokens—ranks third on benchmarks while Muse Spark 1.1 undercuts everyone at $1.25. Meanwhile, an OpenAI autonomous agent accidentally hacked Hugging Face, then Hugging Face used the open GLM 5.2 model to analyze the attack logs when proprietary systems refused, proving that openness can solve security problems closed-model providers won't touch. Google's Gemini 3.5 Flash costs thirty times more than its predecessor at $1.50 per token, erasing the price advantage "Flash" once promised developers. The result: open weights are commodifying the inference layer while pushing proprietary providers into defensive pricing and aggressive regulation lobbying.
Read the full briefing →