Third-party cyber evaluations involving OpenAI models
openai.com·14h ago
Data Science
Towards AI·23d ago·5 min read
GPT-5.6: What Actually Changed on Your Bill
TL;DR
With the transition from GPT-5.5 to GPT-5.6, the pricing structure remains stable, but several previously free features now incur costs. Notably, cache writes, which were free in 5.5, now cost 1.25 times the input rate in 5.6.
✦ Why It Matters
Evaluate your caching strategy today to avoid increased costs with the new pricing model in GPT-5.6.
Key Takeaways
- Review your current caching practices to identify potential cost increases.
- Monitor token usage closely to manage expenses effectively.
- Consider alternative strategies to reduce reliance on caching.
- Stay informed about updates to pricing structures for future planning.
Full Summary
The migration from GPT-5.5 to GPT-5.6 introduces significant changes in cost structure while maintaining the same flagship pricing of five dollars in and thirty out per million tokens. Previously free cache writes now incur a charge of 1.25 times the input rate, altering the financial landscape for users who relied on caching for efficiency.
This change indicates that caching is no longer a risk-free operation, prompting users to reconsider their caching strategies. The implications of these changes are critical for engineers and researchers who need to optimize their usage to avoid unexpected costs.
By analyzing usage patterns and adjusting caching practices, users can mitigate the impact of these new charges. Overall, the update emphasizes the importance of understanding underlying costs in AI model usage.