💻 Open Models Tack Toward the Frontier
Tom Tunguz argues open-weight AI models keep catching closed frontier models on Chatbot Arena Elo but never durably lead — closed models like GPT-5.2-class systems and Opus keep resetting the frontier. A wave of new open releases just landed: Moonshot's Kimi K3 (2.8T params, July 16), Alibaba's Qwen3.8-Max (2.4T, July 19), DeepSeek V4 exiting preview in mid-July, and Thinking Machines' 975B Apache-2.0 model Inkling (July 15). On pricing, the median open-weight model runs about 15% cheaper than GPT-5.2, with the cheapest (DeepSeek V4 Flash) roughly 90% cheaper — while Anthropic is reportedly about to post its first profitable quarter. The takeaway for investors: this closed-leads/open-catches-up cycle keeps margins and prices competitive across the whole AI market.
🔗 Read more