fixed tracking

src/providers/deepseek.rs

@@ -128,17 +128,36 @@ impl super::Provider for DeepSeekProvider {
         cache_write_tokens: u32,
         registry: &crate::models::registry::ModelRegistry,
     ) -> f64 {
-        helpers::calculate_cost_with_registry(
-            model,
-            prompt_tokens,
-            completion_tokens,
-            cache_read_tokens,
-            cache_write_tokens,
-            registry,
-            &self.pricing,
-            0.14,
-            0.28,
-        )
+        if let Some(metadata) = registry.find_model(model) {
+            if metadata.cost.is_some() {
+                return helpers::calculate_cost_with_registry(
+                    model,
+                    prompt_tokens,
+                    completion_tokens,
+                    cache_read_tokens,
+                    cache_write_tokens,
+                    registry,
+                    &self.pricing,
+                    0.28,
+                    0.42,
+                );
+            }
+        }
+
+        // Custom DeepSeek fallback that correctly handles cache hits
+        let (prompt_rate, completion_rate) = self
+            .pricing
+            .iter()
+            .find(|p| model.contains(&p.model))
+            .map(|p| (p.prompt_tokens_per_million, p.completion_tokens_per_million))
+            .unwrap_or((0.28, 0.42)); // Default to DeepSeek's current API pricing
+
+        let cache_hit_rate = prompt_rate / 10.0;
+        let non_cached_prompt = prompt_tokens.saturating_sub(cache_read_tokens);
+
+        (non_cached_prompt as f64 * prompt_rate / 1_000_000.0)
+            + (cache_read_tokens as f64 * cache_hit_rate / 1_000_000.0)
+            + (completion_tokens as f64 * completion_rate / 1_000_000.0)
     }
 
     async fn chat_completion_stream(

src/providers/gemini.rs

@@ -772,17 +772,36 @@ impl super::Provider for GeminiProvider {
         cache_write_tokens: u32,
         registry: &crate::models::registry::ModelRegistry,
     ) -> f64 {
-        super::helpers::calculate_cost_with_registry(
-            model,
-            prompt_tokens,
-            completion_tokens,
-            cache_read_tokens,
-            cache_write_tokens,
-            registry,
-            &self.pricing,
-            0.075,
-            0.30,
-        )
+        if let Some(metadata) = registry.find_model(model) {
+            if metadata.cost.is_some() {
+                return super::helpers::calculate_cost_with_registry(
+                    model,
+                    prompt_tokens,
+                    completion_tokens,
+                    cache_read_tokens,
+                    cache_write_tokens,
+                    registry,
+                    &self.pricing,
+                    0.075,
+                    0.30,
+                );
+            }
+        }
+
+        // Custom Gemini fallback that correctly handles cache hits (25% of input cost)
+        let (prompt_rate, completion_rate) = self
+            .pricing
+            .iter()
+            .find(|p| model.contains(&p.model))
+            .map(|p| (p.prompt_tokens_per_million, p.completion_tokens_per_million))
+            .unwrap_or((0.075, 0.30)); // Default to Gemini 1.5 Flash current API pricing
+
+        let cache_hit_rate = prompt_rate * 0.25;
+        let non_cached_prompt = prompt_tokens.saturating_sub(cache_read_tokens);
+
+        (non_cached_prompt as f64 * prompt_rate / 1_000_000.0)
+            + (cache_read_tokens as f64 * cache_hit_rate / 1_000_000.0)
+            + (completion_tokens as f64 * completion_rate / 1_000_000.0)
     }
 
     async fn chat_completion_stream(

src/providers/helpers.rs

@@ -261,9 +261,9 @@ pub fn parse_openai_response(resp_json: &Value, model: String) -> Result<Provide
         .or_else(|| usage["prompt_cache_hit_tokens"].as_u64())
         .unwrap_or(0) as u32;
 
-    // DeepSeek reports cache_write as prompt_cache_miss_tokens (tokens written to cache for future use).
-    // OpenAI doesn't report cache_write in this location, but may in the future.
-    let cache_write_tokens = usage["prompt_cache_miss_tokens"].as_u64().unwrap_or(0) as u32;
+    // DeepSeek reports prompt_cache_miss_tokens which are just regular non-cached tokens.
+    // They do not incur a separate cache_write fee, so we don't map them here to avoid double-charging.
+    let cache_write_tokens = 0;
 
     Ok(ProviderResponse {
         content,
@@ -300,9 +300,7 @@ pub fn parse_openai_stream_chunk(
             .or_else(|| u["prompt_cache_hit_tokens"].as_u64())
             .unwrap_or(0) as u32;
 
-        let cache_write_tokens = u["prompt_cache_miss_tokens"]
-            .as_u64()
-            .unwrap_or(0) as u32;
+        let cache_write_tokens = 0;
 
         Some(StreamUsage {
             prompt_tokens,