diff --git a/src/modules/llms/server/gemini/gemini.models.ts b/src/modules/llms/server/gemini/gemini.models.ts index f165b2fc5..099f5be00 100644 --- a/src/modules/llms/server/gemini/gemini.models.ts +++ b/src/modules/llms/server/gemini/gemini.models.ts @@ -65,7 +65,7 @@ const geminiExpFree: ModelDescriptionSchema['chatPrice'] = { }; -// Pricing based on https://ai.google.dev/pricing (Jan 8, 2026) +// Pricing based on https://ai.google.dev/pricing (Jan 9, 2026) const gemini30ProPricing: ModelDescriptionSchema['chatPrice'] = { input: [{ upTo: 200000, price: 2.00 }, { upTo: null, price: 4.00 }], @@ -90,13 +90,13 @@ const gemini30FlashPricing: ModelDescriptionSchema['chatPrice'] = { const gemini25ProPricing: ModelDescriptionSchema['chatPrice'] = { input: [{ upTo: 200000, price: 1.25 }, { upTo: null, price: 2.50 }], output: [{ upTo: 200000, price: 10.00 }, { upTo: null, price: 15.00 }], - cache: { cType: 'oai-ac', read: [{ upTo: 200000, price: 0.31 }, { upTo: null, price: 0.625 }] }, + cache: { cType: 'oai-ac', read: [{ upTo: 200000, price: 0.125 }, { upTo: null, price: 0.25 }] }, }; const gemini25FlashPricing: ModelDescriptionSchema['chatPrice'] = { input: 0.30, // text/image/video; audio is $1.00 but we don't differentiate yet output: 2.50, // including thinking tokens - cache: { cType: 'oai-ac', read: 0.075 }, // text/image/video; audio is $0.25 but we don't differentiate yet + cache: { cType: 'oai-ac', read: 0.03 }, // text/image/video; audio is $0.10 but we don't differentiate yet }; const gemini25FlashPreviewPricing = gemini25FlashPricing; // The latest model based on the 2.5 Flash model. 2.5 Flash Preview is best for large scale processing, low-latency. @@ -104,7 +104,7 @@ const gemini25FlashPreviewPricing = gemini25FlashPricing; // The latest model ba const gemini25FlashLitePricing: ModelDescriptionSchema['chatPrice'] = { input: 0.10, // text/image/video; audio is $0.30 but we don't differentiate yet output: 0.40, // including thinking tokens - cache: { cType: 'oai-ac', read: 0.025 }, // text/image/video; audio is $0.125 but we don't differentiate yet + cache: { cType: 'oai-ac', read: 0.01 }, // text/image/video; audio is $0.03 but we don't differentiate yet }; const gemini25FlashLitePreviewPricing = gemini25FlashLitePricing; // The latest model based on Gemini 2.5 Flash lite optimized for cost-efficiency, high throughput and high quality.