AI is becoming a bargain hunter's market, with a few luxury models on top
Key Takeaways
- ▸Inference has become a commodity: GPT-4 equivalent capability costs 55x less than 2022, with DeepSeek's pricing triggering a 97% market discount and overnight repricing
- ▸Frontier models command luxury pricing: OpenAI, Google, and Anthropic are raising prices on cutting-edge reasoning models while commodity models approach zero cost
- ▸Enterprise costs surging despite cheaper inference: Companies face 10x cost increases due to metered pricing, longer agentic workflows, and heavy frontier model usage for complex tasks
Summary
The artificial intelligence market is fracturing into two distinct segments: commodity inference models experiencing dramatic price declines, while frontier models command premium pricing. According to industry analysis, GPT-4-equivalent capability has plummeted from $20 per million tokens in late 2022 to just $0.40 today—a 55x decline. This commoditization was further accelerated when DeepSeek released its R1 reasoning model at $0.55–$2.19 per million tokens in January 2025, undercutting OpenAI's o1-preview ($15–$60) by 97 percent and triggering an immediate market repricing.
However, as inference becomes cheaper, frontier model pricing has surged in the opposite direction. OpenAI doubled GPT-5.5 pricing to $5–$30 per million tokens, while Google's Gemini Flash 3.5 arrived three to six times more expensive than its predecessor. Anthropic's recent Claude Sonnet 5 release continues this trend, despite lower per-token costs, by consuming more tokens to produce equivalent results. Anthropic also shifted corporate customers from per-seat pricing to metered pricing with restricted use of subsidized plans.
The shift toward metered pricing and longer, agentic tasks has driven dramatic cost increases for enterprises, with some companies seeing AI costs jump 10x since January 2026. Industry measurement platforms report that companies are now spending 10–20% of labor costs on tokens, yet higher spending doesn't necessarily correlate with productivity gains. Early data suggests an inflection point around 35–40% of client spending where additional token consumption fails to boost output, indicating the market is entering a phase of intelligent optimization rather than reckless scaling.
- Open-weight models pose competitive pressure: Models like Kimi 2.6/2.7 and GLM 5.2 are approaching Anthropic Opus capability at fractions of the cost, forcing proprietary vendors to defend premium pricing
Editorial Opinion
The emergence of a two-tier AI market reflects genuine industry maturation. Commodity inference commoditizing at near-zero cost is expected and healthy; frontier capabilities justifiably commanding premiums for their reasoning and complexity handling makes economic sense. The real challenge isn't price fragmentation—it's that enterprises lack visibility into where frontier models deliver actual value. Early data showing that AI spending above a certain threshold decouples from productivity gains suggests companies are beginning to ask harder questions about ROI, moving from unconstrained scaling to disciplined optimization.
