OpenAI Cuts AI Prices as the Race Shifts From Token Cost to Successful Outcomes

Artificial intelligence is getting cheaper, but the most important number may no longer be the price of a token.

OpenAI says Luna input pricing has fallen by 80% to $0.20 per million tokens, with output at $1.20. Terra pricing has dropped 20% to $2 for input and $12 for output. The company presents these reductions as part of a broader effort to make intelligence more abundant.

Efficiency is becoming a full-stack competition

Price cuts are possible because AI economics involve much more than the model itself. Chips, networking, data centres, inference software, model design and product behaviour all affect the final cost.

OpenAI says its Sol model helped reduce serving cost by 20%, while speculative-decoding improvements added about 15% efficiency. It also points to reasoning and context changes that improved an ARC-AGI-3 score from 13.3 to 38.3 while using six times fewer output tokens.

Cheap tokens can still be expensive

A low-cost model is not economical if it produces an answer that must be rewritten, checked repeatedly or discarded. For a business, the more useful measurement is cost per successful outcome: the total price of getting a reliable customer response, completed analysis, functioning piece of code or approved marketing asset.

• Measure accuracy and completion, not token price alone.
• Include human review time in the real cost.
• Route routine work to cheaper models and reserve premium reasoning for difficult tasks.

What this means for small businesses

Lower prices can make AI-powered services affordable for smaller organisations. A company may be able to analyse more support conversations, create personalised materials or automate back-office work without enterprise-scale budgets.

But the basic rules remain: protect confidential data, verify important facts and keep accountable people in the decision loop. Affordable intelligence is useful only when it is deployed responsibly.

OpenAI says its systems now reach more than one billion active users and over two million businesses. At that scale, even modest efficiency gains can reshape the market. The next competitive advantage may belong to the provider that delivers not merely the cheapest model, but the most dependable result for every pound spent.

Source: OpenAI’s official “Building abundant intelligence” announcement.