OpenAI aims to enhance AI quality while lowering prices, says Sam Altman

OpenAI aims to enhance AI quality while lowering prices, says Sam Altman

The company's aggressive pricing strategy is turning cheaper models into a revenue growth engine through sheer volume

OpenAI is doing something that sounds paradoxically simple for a company burning through billions in compute costs: making its AI models better and cheaper at the same time. CEO Sam Altman has laid out a vision where the company delivers what he calls “the greatest and the cheapest” in AI.

The strategy is already producing tangible results. On July 30, 2026, OpenAI slashed API pricing for its GPT-5.6 family of models, cutting the Luna model’s costs by 80% and Terra’s by 20%. Luna now runs at $0.20 per million input tokens and $1.20 per million output tokens. Terra sits at $2 and $12 respectively.

Advertisement

The volume play is working

After the price reductions, Luna consumption surged roughly 14-fold, while the effective price dropped by about 10-fold. The math works out to approximately a 34% increase in Luna revenue, even though each individual API call costs a fraction of what it used to.

Altman explained the logic during a late July 2026 podcast appearance, noting that OpenAI is focused on distilling its larger models into smaller, cheaper versions. The goal is maintaining competitive performance while dramatically reducing the cost to serve each request. In certain benchmarks, the company has achieved efficiency gains of 54% fewer output tokens and over 20% reduction in serving costs.

September brought another round of cuts

OpenAI didn’t stop with the July reductions. By September 2026, the company announced further price cuts for its newer GPT-6 model variants. These were priced at roughly half what the previous series cost, a reduction the company attributed to improvements in caching systems and inference optimization.

The competitive context matters here. OpenAI’s pricing moves came as models like Kimi K3 began applying serious pressure on the cost-performance frontier.

OpenAI appears to be measuring its own success through what it calls “Useful Intelligence per Dollar,” a metric that shifts the conversation away from raw benchmark scores and toward practical value delivered relative to cost.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.
OpenAI aims to enhance AI quality while lowering prices, says Sam Altman
OpenAI aims to enhance AI quality while lowering prices, says Sam Altman

The company's aggressive pricing strategy is turning cheaper models into a revenue growth engine through sheer volume

OpenAI is doing something that sounds paradoxically simple for a company burning through billions in compute costs: making its AI models better and cheaper at the same time. CEO Sam Altman has laid out a vision where the company delivers what he calls “the greatest and the cheapest” in AI.

The strategy is already producing tangible results. On July 30, 2026, OpenAI slashed API pricing for its GPT-5.6 family of models, cutting the Luna model’s costs by 80% and Terra’s by 20%. Luna now runs at $0.20 per million input tokens and $1.20 per million output tokens. Terra sits at $2 and $12 respectively.

Advertisement

The volume play is working

After the price reductions, Luna consumption surged roughly 14-fold, while the effective price dropped by about 10-fold. The math works out to approximately a 34% increase in Luna revenue, even though each individual API call costs a fraction of what it used to.

Altman explained the logic during a late July 2026 podcast appearance, noting that OpenAI is focused on distilling its larger models into smaller, cheaper versions. The goal is maintaining competitive performance while dramatically reducing the cost to serve each request. In certain benchmarks, the company has achieved efficiency gains of 54% fewer output tokens and over 20% reduction in serving costs.

September brought another round of cuts

OpenAI didn’t stop with the July reductions. By September 2026, the company announced further price cuts for its newer GPT-6 model variants. These were priced at roughly half what the previous series cost, a reduction the company attributed to improvements in caching systems and inference optimization.

The competitive context matters here. OpenAI’s pricing moves came as models like Kimi K3 began applying serious pressure on the cost-performance frontier.

OpenAI appears to be measuring its own success through what it calls “Useful Intelligence per Dollar,” a metric that shifts the conversation away from raw benchmark scores and toward practical value delivered relative to cost.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.