How much prompt caching saves.
Cached input is 75% to 98% cheaper, depending on the model.
Cached input rates from official pages, as a share of normal input price:
- Claude Fable 5.1: $0.25 vs $10.00 (98% off)
- Claude Sonnet 5.5: $0.10 vs $2.00 (95% off)
- GPT-6.1 Sol: $0.10 vs $2.00 (95% off)
- GPT-6 Astra: $1.00 vs $10.00 (90% off)
- Gemini 3.8 Flash: $0.075 vs $0.75 (90% off)
- DeepSeek Flash (V4.1): $0.006 vs $0.30 (98% off)
- Grok 4.7: $0.50 vs $2.00 (75% off)
Anthropic also charges for cache writes (1.25x base input for the 5-minute cache), so caching pays off once a prefix is reused.
All prices from official provider pages, checked Oct 20, 2026. See sources.
Sponsored
Your API or dev tool here. Reach developers comparing model costs right now.