DistantNews
Support us
🇰🇷 South Korea /Technology

AI models vie for 'cost-effectiveness' as businesses balk at token fees

From Hankyoreh · () Korean

Translated from Korean, summarized and contextualized by DistantNews.

At a glance

News Sources not specified Context piece
  • AI developers are increasingly emphasizing cost-efficiency alongside performance in their latest models due to rising token usage costs for businesses.
  • Companies like OpenAI, Meta, and SpaceX are launching new models with tiered pricing and highlighting lower costs per token compared to competitors.
  • This shift reflects a growing concern among businesses about the substantial expenses of AI integration, prompting a focus on optimizing token usage and exploring cost-saving measures like in-house chip development.

The race to develop cutting-edge artificial intelligence is no longer just about raw power; cost-effectiveness has become a critical battleground. AI developers are now highlighting "more intelligence per token, stronger performance per dollar" as a key selling point for their latest models.

OpenAI, for instance, launched its new "GPT-5.6" with distinct tiers – Sol, Tera, and Luna – emphasizing not only performance but also significantly lower costs compared to previous versions. Tera and Luna models are noted for being cheaper than GPT-5.5 while offering similar or better performance. Meta followed suit, releasing its "Muse Spark 1.1" with a developer-focused pricing model that is substantially cheaper than OpenAI's offerings.

This year is the first year AI costs have become a topic of discussion. Everyone is asking what we can do to help them reduce spending or increase value.

— Sam AltmanOpenAI CEO Sam Altman discussed the growing importance of AI cost-efficiency in a CNBC interview.

SpaceX AI's "Grok 4.5," released just before its competitors, also claims double the token efficiency of leading rival models. This intense competition stems from the escalating token costs businesses are now facing. Initially, companies encouraged widespread AI use, treating it as a productivity metric. However, the reality of hefty monthly bills has shifted the focus to efficient token management.

The AI models have new standards for both intelligence and efficiency.

— OpenAIOpenAI described its new Sol model within the GPT-5.6 lineup.

Companies like Uber have already implemented cost controls after exhausting their annual AI coding tool budget in just four months. Some U.S. firms are even exploring Chinese AI models for their cost-effectiveness. OpenAI CEO Sam Altman acknowledged this trend, stating that "this year is the first year AI costs have become a topic of discussion," with many clients seeking ways to reduce spending or increase value.

In response to these cost pressures, developers are also investing in in-house chip development to reduce reliance on external providers like Nvidia and lower operational expenses. Meta plans to begin mass-producing its custom AI chip "Iris" in September, while OpenAI has unveiled a prototype inference chip co-developed with Broadcom. This strategic move aims to control costs associated with the massive computations required for AI training and inference.

The Tera and Luna models offer superior or similar performance to GPT-5.5 but are significantly cheaper.

— OpenAIOpenAI highlighted the cost-effectiveness of its Tera and Luna models within the GPT-5.6 lineup.
DistantNews Editorial

Originally published by Hankyoreh in Korean. Translated, summarized, and contextualized by our editorial team with added local perspective. Read our editorial standards.

More Stories

From Our Blog