AI Price War: Anthropic and OpenAI Push Model Costs Lower

AI model prices are falling as competition between the major providers increases. Anthropic launched Claude Opus 5.5 this week and says it costs around 40% less to run than Opus 5 for typical workloads. Its standard API price is now $4 per million input tokens and $20 per million output tokens, compared with $5 and $25 for Opus 5. At the same time, OpenAI’s GPT-6 Sol costs $2 input / $10 output, Google’s Gemini 3.1 Pro costs $2 / $12, and xAI’s Grok 4.6 costs $2 / $6 per million tokens. This puts Anthropic’s price reduction into perspective: Claude Opus has become cheaper, but its headline token price remains above the competing models shown below.

The bigger story is that AI companies are increasingly competing on cost as well as intelligence. Anthropic’s 40% figure is particularly interesting because it refers to the estimated cost of completing typical workloads, not simply the reduction in its listed token price. For businesses, this means comparing models based only on price per token may no longer be enough. The important measure is becoming how much useful work a model can complete for each dollar spent. As models become more efficient and competition increases, powerful AI could continue to become cheaper to use, opening the door to more AI applications at scale.