PriceIndex

AI model capabilities

AI models that support prompt caching

Compare models with listed implicit prompt caching and standard token prices. Compare 222 supporting models by input price, output price, context window and published intelligence. Find the cheapest option for a reference text workload, then check the serving API before you build.

222 supporting models31 AI providersUSD · prices per 1M tokens

Lowest reference cost

Ling 3.0 Flash

$0.04

1M input + 250K output tokens. Tool fees and discounts excluded.

Highest published intelligence

Claude Opus 5.5

57.6 AA Intelligence Index

Among supporting models with a published score; not a capability-specific benchmark.

Cheapest AI models with prompt caching

Ranked by the cost of 1 million input and 250,000 output tokens at standard rates. A different token mix can change the cheapest model. Support requires an explicitly listed catalog feature, not an inference from a model family.

GLM 5.3 FlashZ.ai
Input /1M
$0.03
Output /1M
$0.08
Reference cost
$0.05
Context
1M
AA Intelligence
41.8
gpt-oss-20bOpenAI
Input /1M
$0.03
Output /1M
$0.13
Reference cost
$0.06
Context
131K
AA Intelligence
9
Qwen3.7 FlashQwen
Input /1M
$0.03
Output /1M
$0.13
Reference cost
$0.06
Context
1M
AA Intelligence
—
Solar Mini 4Upstage
Input /1M
$0.05
Output /1M
$0.20
Reference cost
$0.10
Context
524K
AA Intelligence
—

Showing 15 of 222

222 matching models · Missing intelligence scores are shown as —.

What is prompt caching?

Prompt caching can reuse eligible repeated input rather than process the entire prompt again. This directory tracks explicitly listed implicit caching support. Explicit caches may exist on other APIs and are not inferred here. Cache-hit rates, minimum prompt sizes, retention and prices depend on the provider.

Before choosing a model

  • Check minimum cacheable input length and retention rules.
  • Use the provider’s cached-input rate when estimating verified cache hits.
  • Measure cache-hit usage; standard token prices do not assume a discount.

Estimate your actual workload

Use your request volume and prompt and response sizes to compare monthly text-token costs. Add any search, external tool or provider-specific charges separately. Standard token prices do not assume cache hits or batch discounts.

LLM token cost calculator Pricing methodology

Prompt caching FAQs

222 active models with positive input and output prices currently list prompt caching in the PriceIndex catalog. The table lists every match. A model is included only when its tracked feature data explicitly lists this capability; missing data is not treated as support.

Ling 3.0 Flash has the lowest estimated text-token cost in this list for 1 million input tokens and 250,000 output tokens: $0.0368. A different input/output mix can change the ranking. This excludes tool fees, caching, discounts and taxes.

No. Capability availability can differ between serving APIs and model versions. Follow a model’s API documentation link and verify the exact version, endpoint and settings before choosing it.

We use standard USD input and output token prices: input price × 1 million tokens plus output price × 250,000 tokens, divided by 1 million. The directory excludes retired models, missing prices and zero-priced entries. Published AA Intelligence Index scores are shown when available and never estimated.

Check minimum cacheable input length and retention rules. Use the provider’s cached-input rate when estimating verified cache hits. Measure cache-hit usage; standard token prices do not assume a discount.

Explore more AI capabilities