fix: count OpenRouter cache writes and drop unverifiable rates

backend/windmill-ai/src/ai_types.rs

@@ -190,6 +190,12 @@ pub const MAX_MODEL_RATE: f64 = 1000.0;
 /// which never deserializes it into `AIConfig`, so the typed check on the
 /// workspace path does not cover it.
 pub fn validate_model_pricing_json(ai_config: &serde_json::Value) -> Result<(), String> {
+    // The container itself has to be checked too: a non-object `ai_config` persists
+    // here and then fails to deserialize as `AIConfig`, which drops the whole
+    // instance config back to its default for every workspace inheriting it.
+    if !ai_config.is_null() && !ai_config.is_object() {
+        return Err("ai_config must be an object".to_string());
+    }
     let pricing = match ai_config.get("model_pricing") {
         None | Some(serde_json::Value::Null) => return Ok(()),
         // A present-but-wrong shape must be rejected, not skipped: it would persist

frontend/src/lib/components/copilot/chat/tokenUsage.ts

@@ -193,7 +193,12 @@ export function openAICompletionsUsageToChatTokenUsage(
 				prompt_tokens?: number | null
 				completion_tokens?: number | null
 				total_tokens?: number | null
-				prompt_tokens_details?: { cached_tokens?: number | null } | null
+				prompt_tokens_details?: {
+					cached_tokens?: number | null
+					/** OpenRouter surfaces Anthropic's cache creation here; OpenAI, whose
+					 * caching is automatic and unbilled, reports no such field. */
+					cache_write_tokens?: number | null
+				} | null
 				/** OpenRouter reports what it actually charged when the request opts in. */
 				cost?: number | null
 		  }
@@ -208,7 +213,7 @@ export function openAICompletionsUsageToChatTokenUsage(
 		completion,
 		total: usage?.total_tokens ?? prompt + completion,
 		cacheRead: usage?.prompt_tokens_details?.cached_tokens ?? 0,
-		cacheWrite: 0,
+		cacheWrite: usage?.prompt_tokens_details?.cache_write_tokens ?? 0,
 		...(typeof usage?.cost === 'number' ? { cost: usage.cost } : {})
 	}
 }

frontend/src/lib/components/copilot/modelConfig.ts

@@ -145,11 +145,14 @@ export function buildModelMatchers<T>(
 			// its own price, not another route to this one — so under strictVariants an
 			// entry does not match when a further *name* segment follows. What follows
 			// is only a decoration when it is a date (`-20251101`), Bedrock's `-v1`, or
-			// one of the alias words below (`claude-3-5-haiku-latest` is the same model
-			// as `claude-3-5-haiku`, and is a shipped default). Off by default: for a
-			// context window an inherited value is a safe approximation, for a price it
-			// is a wrong number.
-			strictVariants ? `(?!-(?!v\\d|${DECORATIVE_SUFFIXES.join('|')})[a-z])` : ''
+			// one of the alias words below, at the very end of the id
+			// (`claude-3-5-haiku-latest` is the same model as `claude-3-5-haiku`, and is
+			// a shipped default; `gpt-5-preview-pro` would be a different one again).
+			// Off by default: for a context window an inherited value is a safe
+			// approximation, for a price it is a wrong number.
+			strictVariants
+				? `(?!-(?!(?:v\\d|${DECORATIVE_SUFFIXES.join('|')})$)[a-z])`
+				: ''
 		].join('')
 		return [new RegExp(pattern + guards), value]
 	})

frontend/src/lib/components/copilot/modelPricing.test.ts

@@ -53,8 +53,10 @@ describe('resolveModelPrice', () => {
 		expect(
 			resolveModelPrice('openrouter', '~anthropic/claude-sonnet-latest', undefined)?.price.input
 		).toBe(3)
-		// …while a genuine sub-model stays unpriced.
+		// …while a genuine sub-model stays unpriced, including one hiding behind a
+		// decoration.
 		expect(resolveModelPrice('openai', 'gpt-5-pro', undefined)).toBeUndefined()
+		expect(resolveModelPrice('openai', 'gpt-5-preview-pro', undefined)).toBeUndefined()
 	})
 
 	it('prefers a workspace override, keeping the model’s own cache ratios', () => {

frontend/src/lib/components/copilot/modelPricing.ts

@@ -110,13 +110,13 @@ const MODEL_PRICES: [name: string, price: PriceEntry | null][] = [
 	['o4-mini', { input: 1.1, output: 4.4, cacheRead: 0.275 }],
 	['o3-mini', { input: 1.1, output: 4.4, cacheRead: 0.55 }],
 	['o3', { input: 2, output: 8, cacheRead: 0.5 }],
-	// Google — a cached read is a quarter of input across the 2.5 family; the 3.x
-	// families are not tracked
+	// Google — left unpriced. Gemini's cached-input rates do not follow a single
+	// ratio and its Pro tier charges more above a context threshold, which a flat
+	// per-model rate cannot express; an estimate here would be wrong in a direction
+	// nobody can see. A workspace that runs Gemini sets its own rates.
 	['gemini-3.1', null],
 	['gemini-3', null],
-	['gemini-2.5-flash-lite', { input: 0.1, output: 0.4, cacheRead: 0.025 }],
-	['gemini-2.5-flash', { input: 0.3, output: 2.5, cacheRead: 0.075 }],
-	['gemini-2.5-pro', { input: 1.25, output: 10, cacheRead: 0.31 }]
+	['gemini-2.5', null]
 ]
 
 const MODEL_PRICE_MATCHERS = buildModelMatchers(