The AI industry is experiencing a shift toward more affordable mid-tier models. While flagship models like Anthropic's Fable 5.1 and OpenAI's GPT 5.6 Sol remain the smartest, they are significantly more expensive. For example, Claude Fable 5.1 costs $3.69 per task on the Intelligence Index test, while Gemini 3.8 Flash costs $0.58, offering 90% of the flagship's capability at roughly one-sixth of the price.
Competition is intensifying in the mid-range segment. Meta's Muse Spark 1.3 (xhigh) currently holds the title for the best intelligence-to-price ratio, with a score of 61 and a cost of $0.55 per task. Other competitive options include Kimi K3 (max) at $0.84 per task with a score of 60, and Deepseek V4 Pro at $0.27 per task with a score of 53.
The demand for affordable intelligence is driving a massive surge in token usage. According to OpenRouter, paid token use has increased 25-fold over the past year and doubled in the last month. This growth is primarily driven by mid-tier models. OpenAI's GPT 5.6 Luna, the most used model, processed nearly 12 trillion tokens in the past month, while Z-Ai's GLM 5.3 Flash saw a more than 1,000% month-over-month increase, processing 11.4 trillion tokens.
The economic reality of the AI industry is that users are willing to pay significantly less for intelligence. High-end models like Fable 5 and GPT 5.6 Sol are used far less frequently, with Fable 5's usage in the low billions of tokens and GPT 5.6 Sol at 1.8 trillion monthly tokens. This trend suggests a move toward efficiency and lower costs as hardware for inference grows more powerful.
