AI pricing benchmarks show mid-tier models delivering 90% capability at one-sixth cost

Token usage surged 25x over past year, with mid-tier AI models now providing flagship capability at one-sixth the cost

Token usage for AI models increased 25 times over the past year, doubling within the previous month. Current benchmark pricing reveals mid-tier models deliver 90% of flagship capability at one-sixth the cost—specifically, Kimi K3 (low) costs $0.24 per task versus $0.84 for Kimi K3 (max), while Deepseek V4 Pro ($0.27) and Gemini 3.8 Flash ($0.58) offer significant cost reductions relative to flagship models. The Intelligence Index benchmark shows these price points represent real-world pricing for specific model tiers, though real-world performance may differ from benchmark results.

This pricing shift reduces barriers to high-capacity AI access for users who require substantial capability without flagship model costs. The Pareto frontier ranking for Gemini 3.8 Flash (held for 3.5 hours on September 2, 2026) highlights how rapidly competitive pricing evolves. However, benchmark gaming during model design and the time-sensitive nature of these rankings mean real-world adoption may not match current pricing claims. For now, the cost gap between mid-tier and flagship models remains the most significant accessibility factor in AI pricing.

Source: Tom's Hardware