Merge pull request #85 from ekoeppen/vertex-models

Update Google Gemini Vertex models

commit: 4d6e4be641c17bd1972d7e8fbc9f7d6f08343c8b
author: Patric Eckhart · 3mo ago · 2026-07-18 08:19 UTC
committer: GitHub · 3mo ago · 2026-07-18 08:19 UTC
merge of: 6a0d7ca + 197d598 · diff against the first parent
signature: GPG UnverifiedB5690EEEBB952194 The key is unknown
refs: tag: v0.2.91

2 files · +14 -7

M packages/provider/catalog_builtin.go +2 -7


@@ -173,19 +173,14 @@ var builtinCatalog = []Model{
173	173		{Provider: "google", ID: "gemma-4-26b-a4b-it", DisplayName: "Gemma 4 26B A4B IT", ContextWindow: 262144, MaxOutput: 32768, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
174	174		{Provider: "google", ID: "gemma-4-31b-it", DisplayName: "Gemma 4 31B IT", ContextWindow: 262144, MaxOutput: 32768, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
175	175		// ----- google-vertex -----
176		-	{Provider: "google-vertex", ID: "gemini-1.5-flash", DisplayName: "Gemini 1.5 Flash (Vertex)", ContextWindow: 1000000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0.01875, BaseURL: "https://{location}-aiplatform.googleapis.com"},
177		-	{Provider: "google-vertex", ID: "gemini-1.5-flash-8b", DisplayName: "Gemini 1.5 Flash-8B (Vertex)", ContextWindow: 1000000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.0375, PriceOutput: 0.15, PriceCacheRead: 0.01, BaseURL: "https://{location}-aiplatform.googleapis.com"},
178		-	{Provider: "google-vertex", ID: "gemini-1.5-pro", DisplayName: "Gemini 1.5 Pro (Vertex)", ContextWindow: 1000000, MaxOutput: 8192, Reasoning: false, PriceInput: 1.25, PriceOutput: 5, PriceCacheRead: 0.3125, BaseURL: "https://{location}-aiplatform.googleapis.com"},
179		-	{Provider: "google-vertex", ID: "gemini-2.0-flash", DisplayName: "Gemini 2.0 Flash (Vertex)", ContextWindow: 1048576, MaxOutput: 8192, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0.0375, BaseURL: "https://{location}-aiplatform.googleapis.com"},
180		-	{Provider: "google-vertex", ID: "gemini-2.0-flash-lite", DisplayName: "Gemini 2.0 Flash Lite (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0.01875, BaseURL: "https://{location}-aiplatform.googleapis.com"},
181	176		{Provider: "google-vertex", ID: "gemini-2.5-flash", DisplayName: "Gemini 2.5 Flash (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.3, PriceOutput: 2.5, PriceCacheRead: 0.03, BaseURL: "https://{location}-aiplatform.googleapis.com"},
182	177		{Provider: "google-vertex", ID: "gemini-2.5-flash-lite", DisplayName: "Gemini 2.5 Flash Lite (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.4, PriceCacheRead: 0.01, BaseURL: "https://{location}-aiplatform.googleapis.com"},
183		-	{Provider: "google-vertex", ID: "gemini-2.5-flash-lite-preview-09-2025", DisplayName: "Gemini 2.5 Flash Lite Preview 09-25 (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.4, PriceCacheRead: 0.01, BaseURL: "https://{location}-aiplatform.googleapis.com"},
184	178		{Provider: "google-vertex", ID: "gemini-2.5-pro", DisplayName: "Gemini 2.5 Pro (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: "https://{location}-aiplatform.googleapis.com"},
185	179		{Provider: "google-vertex", ID: "gemini-3-flash-preview", DisplayName: "Gemini 3 Flash Preview (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0.05, BaseURL: "https://{location}-aiplatform.googleapis.com"},
186	180		{Provider: "google-vertex", ID: "gemini-3-pro-preview", DisplayName: "Gemini 3 Pro Preview (Vertex)", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://{location}-aiplatform.googleapis.com"},
	181	+	{Provider: "google-vertex", ID: "gemini-3.1-flash-lite", DisplayName: "Gemini 3.1 Flash Lite (Vertex)", ContextWindow: 1048576, MaxOutput: 65535, Reasoning: true, PriceInput: 0.25, PriceOutput: 1.5, PriceCacheRead: 0.025, BaseURL: "https://{location}-aiplatform.googleapis.com"},
187	182		{Provider: "google-vertex", ID: "gemini-3.1-pro-preview", DisplayName: "Gemini 3.1 Pro Preview (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://{location}-aiplatform.googleapis.com"},
188		-	{Provider: "google-vertex", ID: "gemini-3.1-pro-preview-customtools", DisplayName: "Gemini 3.1 Pro Preview Custom Tools (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://{location}-aiplatform.googleapis.com"},
	183	+	{Provider: "google-vertex", ID: "gemini-3.5-flash", DisplayName: "Gemini 3.5 Flash (Vertex)", ContextWindow: 1048576, MaxOutput: 65535, Reasoning: true, PriceInput: 1.5, PriceOutput: 9, PriceCacheRead: 0.15, BaseURL: "https://{location}-aiplatform.googleapis.com"},
189	184		// ----- groq -----
190	185		{Provider: "groq", ID: "deepseek-r1-distill-llama-70b", DisplayName: "DeepSeek R1 Distill Llama 70B", ContextWindow: 131072, MaxOutput: 8192, Reasoning: true, PriceInput: 0.75, PriceOutput: 0.99, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
191	186		{Provider: "groq", ID: "gemma2-9b-it", DisplayName: "Gemma 2 9B", ContextWindow: 8192, MaxOutput: 8192, Reasoning: false, PriceInput: 0.2, PriceOutput: 0.2, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},

 	{Provider: "google", ID: "gemma-4-26b-a4b-it", DisplayName: "Gemma 4 26B A4B IT", ContextWindow: 262144, MaxOutput: 32768, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
 	{Provider: "google", ID: "gemma-4-31b-it", DisplayName: "Gemma 4 31B IT", ContextWindow: 262144, MaxOutput: 32768, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
 	// ----- google-vertex -----
-	{Provider: "google-vertex", ID: "gemini-1.5-flash", DisplayName: "Gemini 1.5 Flash (Vertex)", ContextWindow: 1000000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0.01875, BaseURL: "https://{location}-aiplatform.googleapis.com"},
-	{Provider: "google-vertex", ID: "gemini-1.5-flash-8b", DisplayName: "Gemini 1.5 Flash-8B (Vertex)", ContextWindow: 1000000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.0375, PriceOutput: 0.15, PriceCacheRead: 0.01, BaseURL: "https://{location}-aiplatform.googleapis.com"},
-	{Provider: "google-vertex", ID: "gemini-1.5-pro", DisplayName: "Gemini 1.5 Pro (Vertex)", ContextWindow: 1000000, MaxOutput: 8192, Reasoning: false, PriceInput: 1.25, PriceOutput: 5, PriceCacheRead: 0.3125, BaseURL: "https://{location}-aiplatform.googleapis.com"},
-	{Provider: "google-vertex", ID: "gemini-2.0-flash", DisplayName: "Gemini 2.0 Flash (Vertex)", ContextWindow: 1048576, MaxOutput: 8192, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0.0375, BaseURL: "https://{location}-aiplatform.googleapis.com"},
-	{Provider: "google-vertex", ID: "gemini-2.0-flash-lite", DisplayName: "Gemini 2.0 Flash Lite (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0.01875, BaseURL: "https://{location}-aiplatform.googleapis.com"},
 	{Provider: "google-vertex", ID: "gemini-2.5-flash", DisplayName: "Gemini 2.5 Flash (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.3, PriceOutput: 2.5, PriceCacheRead: 0.03, BaseURL: "https://{location}-aiplatform.googleapis.com"},
 	{Provider: "google-vertex", ID: "gemini-2.5-flash-lite", DisplayName: "Gemini 2.5 Flash Lite (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.4, PriceCacheRead: 0.01, BaseURL: "https://{location}-aiplatform.googleapis.com"},
-	{Provider: "google-vertex", ID: "gemini-2.5-flash-lite-preview-09-2025", DisplayName: "Gemini 2.5 Flash Lite Preview 09-25 (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.4, PriceCacheRead: 0.01, BaseURL: "https://{location}-aiplatform.googleapis.com"},
 	{Provider: "google-vertex", ID: "gemini-2.5-pro", DisplayName: "Gemini 2.5 Pro (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: "https://{location}-aiplatform.googleapis.com"},
 	{Provider: "google-vertex", ID: "gemini-3-flash-preview", DisplayName: "Gemini 3 Flash Preview (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0.05, BaseURL: "https://{location}-aiplatform.googleapis.com"},
 	{Provider: "google-vertex", ID: "gemini-3-pro-preview", DisplayName: "Gemini 3 Pro Preview (Vertex)", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://{location}-aiplatform.googleapis.com"},
+	{Provider: "google-vertex", ID: "gemini-3.1-flash-lite", DisplayName: "Gemini 3.1 Flash Lite (Vertex)", ContextWindow: 1048576, MaxOutput: 65535, Reasoning: true, PriceInput: 0.25, PriceOutput: 1.5, PriceCacheRead: 0.025, BaseURL: "https://{location}-aiplatform.googleapis.com"},
 	{Provider: "google-vertex", ID: "gemini-3.1-pro-preview", DisplayName: "Gemini 3.1 Pro Preview (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://{location}-aiplatform.googleapis.com"},
-	{Provider: "google-vertex", ID: "gemini-3.1-pro-preview-customtools", DisplayName: "Gemini 3.1 Pro Preview Custom Tools (Vertex)", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://{location}-aiplatform.googleapis.com"},
+	{Provider: "google-vertex", ID: "gemini-3.5-flash", DisplayName: "Gemini 3.5 Flash (Vertex)", ContextWindow: 1048576, MaxOutput: 65535, Reasoning: true, PriceInput: 1.5, PriceOutput: 9, PriceCacheRead: 0.15, BaseURL: "https://{location}-aiplatform.googleapis.com"},
 	// ----- groq -----
 	{Provider: "groq", ID: "deepseek-r1-distill-llama-70b", DisplayName: "DeepSeek R1 Distill Llama 70B", ContextWindow: 131072, MaxOutput: 8192, Reasoning: true, PriceInput: 0.75, PriceOutput: 0.99, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
 	{Provider: "groq", ID: "gemma2-9b-it", DisplayName: "Gemma 2 9B", ContextWindow: 8192, MaxOutput: 8192, Reasoning: false, PriceInput: 0.2, PriceOutput: 0.2, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},

M packages/provider/provider_test.go +12 -0


@@ -41,6 +41,18 @@ func TestModelCatalog(t *testing.T) {
41	41		}
42	42		}
43	43
	44	+	func TestVertexGeminiFlashCatalogLimits(t *testing.T) {
	45	+	for _, id := range []string{"gemini-3.1-flash-lite", "gemini-3.5-flash"} {
	46	+	m, err := FindModel("google-vertex", id)
	47	+	if err != nil {
	48	+	t.Fatal(err)
	49	+	}
	50	+	if m.ContextWindow != 1048576 \|\| m.MaxOutput != 65535 \|\| !m.Reasoning {
	51	+	t.Errorf("%s metadata = %+v", id, m)
	52	+	}
	53	+	}
	54	+	}
	55	+
44	56		func TestComputeCost(t *testing.T) {
45	57		m, _ := FindModel("anthropic", "claude-sonnet-4-5")
46	58		cost := ComputeCost(m, Usage{InputTokens: 1_000_000, OutputTokens: 1_000_000})


@@ -41,6 +41,18 @@ func TestModelCatalog(t *testing.T) {
41	}	41		}
42	}	42		}
43		43
		44	+	func TestVertexGeminiFlashCatalogLimits(t *testing.T) {
		45	+	for _, id := range []string{"gemini-3.1-flash-lite", "gemini-3.5-flash"} {
		46	+	m, err := FindModel("google-vertex", id)
		47	+	if err != nil {
		48	+	t.Fatal(err)
		49	+	}
		50	+	if m.ContextWindow != 1048576 \|\| m.MaxOutput != 65535 \|\| !m.Reasoning {
		51	+	t.Errorf("%s metadata = %+v", id, m)
		52	+	}
		53	+	}
		54	+	}
		55	+
44	func TestComputeCost(t *testing.T) {	56		func TestComputeCost(t *testing.T) {
45	m, _ := FindModel("anthropic", "claude-sonnet-4-5")	57		m, _ := FindModel("anthropic", "claude-sonnet-4-5")
46	cost := ComputeCost(m, Usage{InputTokens: 1_000_000, OutputTokens: 1_000_000})	58		cost := ComputeCost(m, Usage{InputTokens: 1_000_000, OutputTokens: 1_000_000})