The SDK ships model ID constants in two static classes in the GenerativeAI namespace:

• GoogleAIModels: Gemini API (Google AI Studio) model IDs, most with the models/ prefix.
• VertexAIModels: Vertex AI model IDs, without a prefix.

You can always pass any model ID string yourself. The constants are a convenience and follow Google's models and deprecations pages as of the v4.0.0 release.

var model = new GenerativeModel(apiKey, GoogleAIModels.Gemini38Flash);

GoogleAIModels.DefaultGeminiModel is Gemini38Flash (models/gemini-3.8-flash). It was gemini-2.5-flash in v3.x. The SDK uses it when you don't pass a model and the GOOGLE_AI_MODEL environment variable is not set.

Vertex AI and the global location: Google serves the newest Gemini models on Vertex AI from the global location. With region: "global", the SDK uses the https://aiplatform.googleapis.com endpoint instead of https://{region}-aiplatform.googleapis.com. If a new model returns "not found" in a regional location, try global:

var vertex = new VertexAI(projectId: "my-project", region: "global");
var model = vertex.CreateGenerativeModel(VertexAIModels.Gemini.Gemini38Flash);

Google deprecated temperature, topP and topK for Gemini 3.6 and later models. The SDK never sets them on its own. The GenerationConfig properties are still there for older models and Vertex AI.

See Multimodal Live API.

All Imagen models are retired, on the Gemini API and (since June 30, 2026) on Vertex AI, so use the Gemini image models above. See Image Generation.

Veo is retired on the Gemini API. VideoGenerationModel now throws NotSupportedException on the Gemini API. To generate video there, use Gemini Omni through the Interactions API, or use Veo on Vertex AI. See Video Generation.

EmbeddingModel checks the model against SupportedEmbedingModels.All. In v4 the check compares bare model IDs, so models/gemini-embedding-001, publishers/google/models/gemini-embedding-001 and gemini-embedding-001 all pass. See Embeddings.

See Interactions API and Agent APIs.

VertexAIModels.Gemini also keeps a few preview IDs (Gemini25FlashPreviewTTS, Gemini25ProPreviewTTS, Gemini25FlashImagePreview, Gemini3FlashPreview, Gemini3ProPreview, Gemini3ProImagePreview). Prefer the GA IDs above. Retired 2.0 and old 2.5 preview IDs were removed in 4.0.1.

On Vertex AI, the SDK sends Gemini and model-as-a-service embedding models to embedContent/batchEmbedContents. It sends other text embedding models, such as text-embedding-005, text-embedding-004 and text-multilingual-embedding-002, to predict.

• VertexAIModels.Gemma: GemmaModel (gemma), CodeGemma (codegemma), PaliGemma (paligemma)
• VertexAIModels.MedLM: MedLMMedium (medlm-medium), MedLMLarge (medlm-large)

These constants still work, but they are marked [Obsolete(..., false)], so using one gives compiler warning CS0618. The replacement is in the warning message.

Constants for models that Google has shut down were removed in version 4.0.1, so code that uses them no longer compiles. Replace them as follows. If you need one of these IDs anyway, pass the model ID as a string.

Gemini API (GoogleAIModels)

Vertex AI (VertexAIModels)

Based on Google's retired models table.

VideoGenerationModel and VertexAI.CreateVideoGenerationModel() now default to Video.Veo31Generate001.