We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Hacker News·Jun 26·5 min read
Why current LLM costs are not sustainable
TL;DR
AI costs are becoming unsustainable for many companies, with Uber exhausting its annual AI budget in just four months. Current frontier models, like GPT 5.5, are expensive to use, costing $5 per million input tokens and $30 per million output tokens.
✦ Why It Matters
Engineers should anticipate changes in AI pricing and adjust budgets and project plans accordingly.
Key Takeaways
- Monitor AI budget usage to avoid overspending.
- Evaluate the cost-effectiveness of different AI models.
- Stay informed about model performance trends and pricing.
- Prepare for potential shifts in AI cost structures.
Full Summary
AI technology is facing a significant cost issue, as evidenced by companies like Uber, which quickly depleted its annual AI budget. Frontier models, such as GPT 5.5, are among the most expensive, with costs reaching $5 per million input tokens and $30 per million output tokens.
Despite ongoing improvements in AI models, the rate of enhancement is slowing, indicating a potential plateau in performance. Additionally, the training data used for these models is largely exhausted, making further improvements challenging.
As competition among AI labs increases and model performance stabilizes, prices are likely to decrease. For instance, Claude Opus 4.8 is priced the same as its predecessor, indicating a shift in pricing strategy.
These trends suggest that engineers and researchers should prepare for a more competitive landscape with potentially lower costs.