The AI price war is intensifying as OpenAI cuts model costs and closes the enterprise spending gap with Anthropic, which is responding with cheaper Claude models.
The AI price war is moving from a battle over benchmark scores to a much simpler question for businesses: how much does useful artificial intelligence cost?
OpenAI has gained ground on Anthropic among companies buying access to advanced models, according to new market data reported by The Wall Street Journal.
Anthropic had previously held a wide lead among the businesses measured.
According to The Wall Street Journal, Anthropic at one point accounted for roughly 75% of model spending among customers in the dataset it examined.
By September, OpenAI had pulled roughly level with Anthropic among about 120,000 businesses tracked through OpenRouter.
AI price war moves beyond raw model performance
The AI price war reflects a change in what corporate customers care about.
The most capable AI model is not automatically the most economical model to deploy across thousands of employees or millions of automated tasks.
Companies also have to consider token prices, caching, speed and how many attempts a model needs to finish a job correctly.
OpenAI has pushed hard on those economics.
GPT-6 Sol is priced at $2 per million standard input tokens and $10 per million output tokens for short-context requests, according to OpenAI’s published API pricing.
GPT-6 Luna goes much lower, at $0.10 per million input tokens and $0.50 per million output tokens.
The AI Decode previously examined the GPT-6 Sol and Luna price cuts, which brought pricing down sharply from OpenAI’s previous generation.
Those reductions made the AI price war harder for rivals to ignore.
Anthropic is cutting Claude costs too
Anthropic is responding with its own efficiency push.
Its new Claude Haiku 5.5 is aimed directly at high-volume workloads such as summarisation, database queries, classification and customer support.
In its Claude Haiku 5.5 announcement, Anthropic said the model costs about 75% less to run on average than Haiku 4.5.
Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens for prompts within its lower pricing range.
Those figures put it directly alongside GPT-6 Luna on headline token pricing.
Anthropic has also reduced the cost of larger models.
Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens, while Anthropic says lower cache-read pricing reduces the cost of many agent workloads by about 20%.
Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens.
The AI price war therefore now covers almost every tier of the market.
Enterprise buyers are becoming more selective
Price reductions matter because business AI usage can become expensive quickly.
A company building one internal chatbot may process millions of tokens.
A business running autonomous coding or customer-service agents can process billions.
A difference of a few dollars per million tokens can turn into a large annual bill.
That gives companies a reason to route different jobs to different models instead of choosing one provider for everything.
A difficult programming task might go to an expensive frontier model. Routine classification or summarisation can go to a smaller, cheaper model.
This routing approach puts even more pressure on the AI price war.
Providers can no longer rely on customers using their flagship model for every request.
They have to compete at several price and capability levels.
OpenAI is gaining ground in business AI
OpenAI’s improved enterprise position is important because Anthropic has built a strong reputation among professional developers, particularly through Claude Code.
The Journal reports that some businesses have become more willing to use OpenAI as its newer models lowered costs and improved coding performance.
OpenAI also has Codex and a growing set of business tools that compete directly with Anthropic’s developer products.
That does not mean Anthropic has lost the market.
Enterprise AI spending can shift quickly when a new model launches, prices change or companies update their data policies.
The AI Decode previously examined Anthropic’s rising valuation, driven partly by expectations that Claude could become a major enterprise platform.
The AI price war now puts those expectations under a tougher financial test.
Cheap AI creates another problem
Falling prices are good for developers and businesses buying AI.
They create a harder equation for the companies selling it.
OpenAI and Anthropic both spend heavily on chips, data centers, model training and research staff.
If the amount charged for each unit of AI work keeps falling, usage has to rise quickly enough to compensate.
The industry is betting that cheaper models will create far more demand.
A company that previously used AI for one expensive task may use it for hundreds of smaller tasks once prices fall enough.
That could expand the overall market even as the price of each individual request declines.
The same pattern happened in cloud computing, storage and internet bandwidth.
Whether AI economics follow the same path remains uncertain because frontier models require enormous infrastructure investment.
AI price war is becoming a test of efficiency
The AI price war is unlikely to be decided by the cheapest token alone.
A model that costs half as much but needs twice as many steps to finish a task may offer little saving.
Businesses increasingly care about the total cost of completing the work.
That includes model price, speed, accuracy and the amount of human review required afterward.
OpenAI and Anthropic are both adjusting their products around that calculation.
OpenAI is using Sol and Luna to push lower-cost GPT-6 capability across different workloads. Anthropic is using its Haiku, Sonnet and Opus lineup to give customers similar choices.
Chinese open-weight models add another source of pricing pressure.
The result is an AI price war where every major provider has to show that its models are worth their operating cost, not simply that they can win a benchmark.
OpenAI has gained ground for now.
The next question is whether lower prices increase usage quickly enough to support the enormous cost of building the models companies are discounting.
