diff --git a/packages/ai/CHANGELOG.md b/packages/ai/CHANGELOG.md
index b6e423b8..aa8e1b08 100644
--- a/packages/ai/CHANGELOG.md
+++ b/packages/ai/CHANGELOG.md
@@ -2,6 +2,10 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Fixed DeepSeek prompt cache hits to be tracked from `prompt_cache_hit_tokens` in OpenAI-compatible usage responses ([#3880](https://github.com/badlogic/pi-mono/issues/3880)).
+
 ## [0.70.6] - 2026-04-28
 
 ### Added
diff --git a/packages/ai/src/providers/openai-completions.ts b/packages/ai/src/providers/openai-completions.ts
index f0e167e9..7af3b897 100644
--- a/packages/ai/src/providers/openai-completions.ts
+++ b/packages/ai/src/providers/openai-completions.ts
@@ -956,12 +956,13 @@ function parseChunkUsage(
 	rawUsage: {
 		prompt_tokens?: number;
 		completion_tokens?: number;
+		prompt_cache_hit_tokens?: number;
 		prompt_tokens_details?: { cached_tokens?: number; cache_write_tokens?: number };
 	},
 	model: Model<"openai-completions">,
 ): AssistantMessage["usage"] {
 	const promptTokens = rawUsage.prompt_tokens || 0;
-	const reportedCachedTokens = rawUsage.prompt_tokens_details?.cached_tokens || 0;
+	const reportedCachedTokens = rawUsage.prompt_tokens_details?.cached_tokens ?? rawUsage.prompt_cache_hit_tokens ?? 0;
 	const cacheWriteTokens = rawUsage.prompt_tokens_details?.cache_write_tokens || 0;
 
 	// Normalize to pi-ai semantics: