fix together provider issues (#3140)

src/backend/drivers/ai-chat/ChatCompletionDriver.ts

@@ -651,8 +651,18 @@ export class ChatCompletionDriver extends PuterDriver {
             // thinking_tokens → output rate fallback
             let rate = costs[key];
             if (typeof rate !== 'number' || !Number.isFinite(rate)) {
-                if (key === 'thinking_tokens' && outputRate !== undefined) {
+                if (isOutputKey(key) && outputRate !== undefined) {
                     rate = outputRate;
+                } else if (!isOutputKey(key)) {
+                    const inputRateRaw = costs[inputKey];
+                    if (
+                        typeof inputRateRaw === 'number' &&
+                        Number.isFinite(inputRateRaw)
+                    ) {
+                        rate = inputRateRaw;
+                    } else {
+                        continue;
+                    }
                 } else {
                     continue;
                 }

src/backend/drivers/ai-chat/providers/together/TogetherAIProvider.test.ts

@@ -433,10 +433,10 @@ describe('TogetherAIProvider.complete non-stream output', () => {
             completion_tokens: 50,
             cached_tokens: 0,
         });
-        // Qwen pricing: input=20, output=20 (per million).
+        // Qwen pricing: input=20, output=20 (per million, dollars from API → ×100 for cents).
         expect(overrides).toMatchObject({
-            prompt_tokens: 100 * 20,
-            completion_tokens: 50 * 20,
+            prompt_tokens: 100 * 20 * 100,
+            completion_tokens: 50 * 20 * 100,
         });
     });
 });
@@ -484,13 +484,13 @@ describe('TogetherAIProvider.complete streaming', () => {
             cached_tokens: 0,
         });
 
-        // Qwen pricing: input=20, output=20.
+        // Qwen pricing: input=20, output=20 (dollars from API → ×100 for cents).
         expect(recordSpy).toHaveBeenCalledTimes(1);
         const [, , prefix, overrides] = recordSpy.mock.calls[0]!;
         expect(prefix).toBe('togetherai:Qwen/Qwen2.5-7B-Instruct-Turbo');
         expect(overrides).toMatchObject({
-            prompt_tokens: 4 * 20,
-            completion_tokens: 2 * 20,
+            prompt_tokens: 4 * 20 * 100,
+            completion_tokens: 2 * 20 * 100,
         });
     });
 });

src/backend/drivers/ai-chat/providers/together/TogetherAIProvider.ts

@@ -75,7 +75,11 @@ export class TogetherAIProvider implements IChatProvider {
                     output_cost_key: 'output',
                     costs: {
                         tokens: 1_000_000,
-                        ...model.pricing,
+                        ...Object.fromEntries(
+                            Object.entries(model.pricing ?? {}).map(
+                                ([k, v]) => [k, (v as number) * 100],
+                            ),
+                        ),
                     },
                     max_tokens: model.context_length ?? 8000,
                 });
@@ -126,9 +130,13 @@ export class TogetherAIProvider implements IChatProvider {
 
         const actor = Context.get('actor');
         const models = await this.models();
+        const modelLower = model.toLowerCase();
         const modelUsed =
-            models.find((m) => [m.id, ...(m.aliases || [])].includes(model)) ||
-            models.find((m) => m.id === this.getDefaultModel())!;
+            models.find((m) =>
+                [m.id, ...(m.aliases || [])].some(
+                    (id) => id.toLowerCase() === modelLower,
+                ),
+            ) || models.find((m) => m.id === this.getDefaultModel())!;
         const modelIdForParams = modelUsed.id.startsWith('togetherai:')
             ? modelUsed.id.slice('togetherai:'.length)
             : modelUsed.id;