AlphaSense said testing of AI models on 246 financial analysis tasks found that OpenAI's GPT-5.6 Sol and Anthropic's Opus 4.8 delivered better answers at lower real-world cost than Chinese rivals Kimi K3 and GLM-5.2, challenging a common view based on headline per-token pricing alone. While Moonshot's Kimi K3 was listed at $15 per 1 million output tokens versus $25 for Opus 4.8 and $30 for GPT-5.6 Sol, the study said higher-capability models often used fewer tokens and fewer processing steps to complete the same work. GPT-5.6 Sol posted a median total cost about 13% lower than Kimi K3 with quality roughly 20% higher, while Opus 4.8 scored around 13% higher at roughly half the cost of Kimi K3. The findings arrive as enterprises weigh frontier AI systems from OpenAI and Anthropic against lower-priced Chinese and open-weight alternatives. AlphaSense CEO Jack Kokko said some models that look expensive on a per-token basis end up costing less because they use tokens more efficiently. The report said the most cost-effective approach may be to evaluate cost per completed task and quality together, and in some cases to route work across multiple models, using stronger systems for planning and smaller ones for execution.