---
id: provider-cache-hit-rate/2026-09
series: provider-cache-hit-rate
title: Prompt cache hit rate by provider, September 2026
period: 2026-09
period_start: 2026-09-01
period_end: 2026-09-30
updated: 2026-11-11
license: CC BY 4.0
canonical: https://www.requesty.ai/data/provider-cache-hit-rate/2026-09
latest: https://www.requesty.ai/data/provider-cache-hit-rate
---

# Prompt cache hit rate by provider, September 2026

> Runware served 93.3% of input tokens from cache in September 2026, the highest among major providers. The median major provider served 75.4%.

Question: What share of input tokens do LLM providers serve from prompt cache?

## Key findings

1. Runware leads at 93.3%; Vertex (Gemini) is lowest at 11.7%.
2. 8 of 17 major providers served at least 80% of input tokens from cache.
3. Nebius moved the most since August 2026: cache hit rate went from 75.9% to 31.8% (-44.1 pp).

## Data

| Provider | Cache hit rate | Traffic share | Organizations (band) |
| --- | --- | --- | --- |
| Runware | 93.30% | 0.96% | 500+ |
| DeepSeek | 90.49% | 4.88% | 500+ |
| Novita | 89.77% | 1.54% | 500+ |
| Sference | 89.25% | 7.88% | 500+ |
| Fireworks | 89.19% | 1.44% | 500+ |
| Vertex (Claude) | 81.63% | 1.77% | 500+ |
| xAI | 80.94% | 0.57% | 100-499 |
| Bedrock | 80.25% | 4.07% | 500+ |
| Anthropic | 75.44% | 7.91% | 500+ |
| Azure | 74.15% | 12.23% | 500+ |
| TensorX | 73.06% | 4.60% | 500+ |
| DeepInfra | 72.51% | 14.72% | 500+ |
| NVIDIA | 68.24% | 0.90% | 500+ |
| OpenAI | 67.71% | 11.36% | 500+ |
| Mistral | 52.95% | 2.95% | 500+ |
| Nebius | 31.79% | 0.52% | 500+ |
| Vertex (Gemini) | 11.74% | 14.84% | 500+ |

## Method

- Cache hit rate is cached input tokens divided by input tokens, the same definition /rankings uses.
- Providers that report no cached tokens are left out of the chart. Only providers with at least 0.5% of published traffic are shown.
- A row is published only when it has at least 20 requests, at least 5 distinct organizations, and no single organization above 80% of its traffic. These are the same floors /rankings uses. Rows below the floors are withheld, not estimated.
- Only shares, rates, medians and organization bands are published. Absolute request, token and spend totals stay private.

## Caveats

- Hit rate depends on workload. Long agent sessions with a stable prefix cache far better than one-off requests.
- Providers report cached tokens differently. Some only count explicit cache reads.

## Cite as

Requesty (2026). Prompt cache hit rate by provider, September 2026. Requesty Data. https://www.requesty.ai/data/provider-cache-hit-rate/2026-09

```bibtex
@misc{requesty_provider_cache_hit_rate_2026_09,
  author       = {{Requesty}},
  title        = {Prompt cache hit rate by provider, September 2026},
  year         = {2026},
  howpublished = {\url{https://www.requesty.ai/data/provider-cache-hit-rate/2026-09}},
  note         = {Requesty Data}
}
```

Source: Requesty production gateway. License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).
Downloads: [JSON](https://www.requesty.ai/data/provider-cache-hit-rate/2026-09/data.json) · [CSV](https://www.requesty.ai/data/provider-cache-hit-rate/2026-09/data.csv) · [Markdown](https://www.requesty.ai/data/provider-cache-hit-rate/2026-09/data.md)