Codex
Cached input counts at a lower rate; cache writes are free.
In the published credit rates, cached input tokens cost a tenth of regular input (GPT-5.6 Sol: 100 credits per million input tokens, 10 for cached input), and Codex does not charge for cache writes. For the 5-hour and weekly allowance included in a plan, OpenAI only says that caching affects usage; it does not publish the discount.
What the official docs say
Charges are based on the model used and the actual input, cached input, and output tokens consumed.Codex does not charge for cache writes.Model choice, context, reasoning, tool use, retrieval, and caching all affect usage, so prompt length alone isn't a reliable estimate.Claude
Cache reads count, at the cached rate; reused project content does not count again.
On claude.ai, project content is cached and only new or uncached parts count when reused. In Claude Code every turn re-reads the whole conversation at the cached rate, so a one-line question in a long session still draws usage for the whole history. The cache lasts an hour on a subscription (five minutes for subagents, and once you are on usage credits); a break longer than that reprocesses your full context. Anthropic does not publish the discount as applied to subscription limits.
What the official docs say
Content in projects is cached and doesn't count against your limits when reused. Similar prompts you use frequently are partially cached.With prompt caching, Claude Code re-reads that history at the cached token rate, so a one-line question in a session that has been open all day still draws usage for the whole conversation.Cache misses: your first message after a break longer than the cache lifetime misses the cache and reprocesses your full context. The lifetime is an hour on a subscription and drops to five minutes once you're drawing on usage creditscache_read_input_tokens — Tokens served from cache on this turn, billed at roughly 10% of the standard input rateIn practice
- The expensive moment is the first message after a long break: the cache has expired and everything is reprocessed.
- Claude Code flags cache misses in /usage when they reach 10% of recent usage.
Quotes checked against the live docs on 2026-09-15.