This catch-up edition covers news from October 20, 2026. We published it on October 23, 2026.
12-hour prompt caching
OpenAI now offers 12-hour prompt caching in the Responses API. The company says the longer retention keeps shared instructions, tools, and context available between requests that arrive hours apart. Source
If your team sends the same context repeatedly, compare the cached-input benefit with the write price: OpenAI charges 2.5 times the input token price for 12-hour cache writes. The announcement includes no workload-specific savings results. Source
Who can use it: Available now in API: v1/responses; OpenAI says you can now choose 12-hour prompt caching.
Keep in mind: This is an announcement, not a measured performance report. The changelog only states that cached context can persist longer and that cache writes cost 2.5x the input token price; it does not provide savings benchmarks or workload-specific results.