@@ -81,6 +81,57 @@ stringData:
8181` ` `
8282{% /callout %}
8383
84+ ## Model Selection Logic
85+
86+ The model selected is generally a waterfall like so.
87+
88+ 1. For a use-case demanding a tool-capable model:
89+
90+ | provider | tool model | default model |
91+ |----------|------------|---------------|
92+ | tool | yes | n/a |
93+ | tool | no | yes |
94+ | default | yes | n/a |
95+ | default | no | yes |
96+
97+
98+ 2. For a high-volume usecase, the tool provider is ignored, and the table above is recognized for just the default provider usecases
99+
100+ 3. For embedding model selection, the logic is:
101+
102+ | provider | embedding model | default model |
103+ |-----------|----------------|---------------|
104+ | embedding | yes | n/a |
105+ | embedding | no | yes |
106+ | default | yes | n/a |
107+ | default | no | yes |
108+
109+ (so very similar to the tool provider use-case)
110+
111+ ## Recommendations
112+
113+ We've found decent success with the following combinations:
114+
115+ For OpenAI:
116+
117+ 1. Any openai model above gpt-4.1-mini, but a tuned setup would chose gpt-4.1 or above for tools, and gpt-4.1-mini for the default
118+ 2. For embeddings, we default to ` text-embedding-3-large`
119+
120+ For VertexAI :
121+
122+ 1. High volume usecases with `gemini-2.5-flash` and tools with `anthropic/claude-sonnet-4-5` is a realistic combo.
123+ 2. You can use OpenAI as your embedding provider, or any of the Vertex embedding model's as well.
124+
125+ For Anthropic, its important to note they have no embedding model, so you'll always have to mix providers. This is a decent setup :
126+
127+ 1. Configure `claude-sonnet-4-5` for tools (set as `toolProvider`)
128+ 2. OpenAI gpt-4.1-mini for volume, which is our default (it's cheaper than most anthropic models) (set as `provider`)
129+ 3. OpenAI `text-embedding-3-large` as your embedding model, which is our default, and would be selected as its the base provider.
130+
131+ {% callout severity="info" %}
132+ Configuring a default model is usually optional, we chose sane defaults for all major providers.
133+ {% /callout %}
134+
84135# # Learn More
85136
86- You can see the full docs for this resource at https://docs.plural.sh/overview/management-api-reference#deploymentsettingsspec
137+ You can see the full docs for this resource at our [Agent API docs]( https://docs.plural.sh/overview/management-api-reference#deploymentsettingsspec)
0 commit comments