Prompt caching AI models: send the same prefix twice, pay once
| # | AI model | Capabilities | API $/1M (in / out) | Context window | Price / mo | Output $/1M | Overview | Visit site |
|---|---|---|---|---|---|---|---|---|
| 81 |  Gemma 4 26B A4B Google |    | $0.07 / $0.34 | 262.1K tokens | API only | 0.34 | View | |
| 82 |  Qwen3.7 Flash Qwen | | $0.03 / $0.13 | 1M tokens | API only | 0.13 | View | |
| 83 | Qwen3.8 Flash Qwen | | $0.15 / $0.47 | 1M tokens | API only | 0.47 | View | |
| 84 |  Kimi K2.6 Moonshot AI |  | $0.95 / $4 | 262.1K tokens | API only | 4 | View | |
| 85 |  Claude Opus 4.6 Anthropic | | $5 / $25 | 1M tokens | API only | 25 | View | |
| 86 |  GPT-5.6 Sol Pro OpenAI | | $2 / $10 | 1.05M tokens | API only | 10 | View | |
| 87 | GPT-6 Astra Pro OpenAI | | $10 / $50 | 1.05M tokens | API only | 50 | View | |
| 88 | Claude Opus 4.7 Anthropic | | $5 / $25 | 1M tokens | API only | 25 | View | |
| 89 | Qwen3.5-Flash Qwen | | $0.065 / $0.26 | 1M tokens | API only | 0.26 | View | |
| 90 | Kimi K2.5 Moonshot AI | | $0.45 / $2.25 | 262.1K tokens | API only | 2.25 | View | |
| 91 | Qwen3.6 35B A3B Qwen | | $0.1 / $0.9 | 262.1K tokens | API only | 0.9 | View | |
| 92 |  GLM 4.7 Z.ai |   | $0.4 / $1.75 | 204.8K tokens | API only | 1.75 | View | |
| 93 | GPT-5.2 OpenAI | | $1.75 / $14 | 400K tokens | API only | 14 | View | |
| 94 | Qwen3.5-35B-A3B Qwen | | $0.3125 / $1.25 | 262.1K tokens | API only | 1.25 | View | |
| 95 | GLM 5 Z.ai |  | $0.6 / $1.92 | 204.8K tokens | API only | 1.92 | View | |
| 96 | Qwen3.5-9B Qwen | | $0.1 / $0.15 | 262.1K tokens | API only | 0.15 | View | |
| 97 |  Grok 4.20 xAI | | $1.25 / $2.5 | 2M tokens | API only | 2.5 | View | |
| 98 | Qwen3 VL 32B Instruct Qwen | | $0.104 / $0.416 | 131.1K tokens | API only | 0.416 | View | |
| 99 |  MiniMax M2.5 MiniMax | | $0.27 / $1.08 | 204.8K tokens | API only | 1.08 | View | |
| 100 | Qwen3 VL 235B A22B Instruct Qwen | | $0.21 / $1.9 | 262.1K tokens | API only | 1.9 | View |
Browse by need
Prompt caching stores an identical prompt prefix between calls, so repeated context is charged at a fraction of the base input rate - typically 10%, and as little as 2.5% on Claude Fable 5.1. If your application sends the same lengthy system prompt with every request, this is usually the biggest saving available to you.
147 models support it. The saving applies to input only, so the effect is largest where prompts are long and answers are short - support bots with detailed instructions, agents with big tool catalogues, RAG systems that resend the same corpus. Cache entries expire, so the economics favour high request frequency against a stable prefix.
Models with this capability
The 147 models below all support this capability. Prices are input and output per 1M tokens from the vendor official price list; for dirhams multiply by 3.6725. Tap a card to open the model page.
How to choose
- Put everything stable at the very front: system prompt, tool definitions, reference documents.
- Keep user-specific content after the cached prefix.
- Check the cache lifetime — short TTLs punish low-traffic applications.
- Account for write cost: some vendors charge a premium to create the entry.
Common mistakes
Who this suits
Related filters
Frequently asked questions
How much does caching save?
What has to stay identical?
When is caching pointless?
How much can I realistically save?
Does caching change answers?
Frequently searched
Rate this page
Help us improve