fix: replace the gpt-5-mini runtime default model before OpenAI retires it

SCORERS.md

@@ -25,7 +25,7 @@ Evaluates whether the output is factually consistent with the expected answer.
 - `input` (string): The input question or prompt
 - `output` (string, required): The generated answer to evaluate
 - `expected` (string, required): The ground truth answer
-- `model` (string, optional): Model to use (default: configured via `init()` or "gpt-5-mini")
+- `model` (string, optional): Model to use (default: configured via `init()` or "gpt-5.6-terra")
 - `client` (Client, optional): Custom OpenAI client
 
 **Score Range:** 0-1
@@ -209,7 +209,7 @@ Evaluates how relevant the retrieved context is to the input question.
 - `input` (string, required): The question
 - `output` (string, required): The generated answer
 - `context` (string[] | string, required): Retrieved context passages
-- `model` (string, optional): Model to use (default: "gpt-5-nano")
+- `model` (string, optional): Model to use (default: "gpt-5.6-luna")
 
 **Score Range:** 0-1
 
@@ -600,7 +600,7 @@ Note: Interpretation varies by scorer type. Binary scorers (ExactMatch, ValidJSO
 
 Many scorers share these common parameters:
 
-- `model` (string): LLM model to use for evaluation (default: configured via `init()` or "gpt-5-mini")
+- `model` (string): LLM model to use for evaluation (default: configured via `init()` or "gpt-5.6-terra")
 - `client` (Client): Custom OpenAI-compatible client
 - `use_cot` (boolean): Enable chain-of-thought reasoning for LLM scorers (default: true)
 - `temperature` (number): LLM temperature setting

js/init-models.test.ts

@@ -36,7 +36,7 @@ describe("init with defaultModel parameter", () => {
 
     expect(getDefaultEmbeddingModel()).toBe("text-embedding-3-large");
     // Completion model should remain at default since we didn't update it
-    expect(getDefaultModel()).toBe("gpt-5-mini");
+    expect(getDefaultModel()).toBe("gpt-5.6-terra");
   });
 
   test("object form can set both models", () => {
@@ -76,7 +76,7 @@ describe("init with defaultModel parameter", () => {
   test("falls back to defaults when not set", () => {
     init();
 
-    expect(getDefaultModel()).toBe("gpt-5-mini");
+    expect(getDefaultModel()).toBe("gpt-5.6-terra");
     expect(getDefaultEmbeddingModel()).toBe("text-embedding-ada-002");
   });
 

js/oai.test.ts

@@ -332,8 +332,8 @@ describe("OAI", () => {
     expect(Object.is(builtClient, otherClient)).toBe(true);
   });
 
-  test("getDefaultModel returns gpt-5-mini by default", () => {
-    expect(getDefaultModel()).toBe("gpt-5-mini");
+  test("getDefaultModel returns gpt-5.6-terra by default", () => {
+    expect(getDefaultModel()).toBe("gpt-5.6-terra");
   });
 
   test("init sets default model", () => {
@@ -346,7 +346,7 @@ describe("OAI", () => {
     expect(getDefaultModel()).toBe("claude-3-5-sonnet-20241022");
 
     init({ defaultModel: undefined });
-    expect(getDefaultModel()).toBe("gpt-5-mini");
+    expect(getDefaultModel()).toBe("gpt-5.6-terra");
   });
 
   test("init can set both client and default model", () => {

js/oai.ts

@@ -203,7 +203,7 @@ export interface InitOptions {
    *
    * Can be either:
    * - A string (for backward compatibility): Sets the default completion model only.
-   *   Defaults to "gpt-5-mini" if not set.
+   *   Defaults to "gpt-5.6-terra" if not set.
    * - An object with `completion` and/or `embedding` properties: Allows setting
    *   default models for different evaluation types. Only the specified models
    *   are updated; others remain unchanged.
@@ -237,7 +237,7 @@ export interface InitOptions {
     | {
         /**
          * Default model for LLM-as-a-judge evaluations (completion).
-         * Defaults to "gpt-5-mini" if not set.
+         * Defaults to "gpt-5.6-terra" if not set.
          */
         completion?: string;
         /**
@@ -300,10 +300,10 @@ export const init = ({ client, defaultModel }: InitOptions = {}) => {
 };
 
 /**
- * Get the configured default completion model, or "gpt-5-mini" if not set.
+ * Get the configured default completion model, or "gpt-5.6-terra" if not set.
  */
 export const getDefaultModel = (): string => {
-  return globalThis.__defaultModel ?? "gpt-5-mini";
+  return globalThis.__defaultModel ?? "gpt-5.6-terra";
 };
 
 /**

py/autoevals/llm.py

@@ -3,7 +3,7 @@
 This module provides a collection of pre-built LLM scorers for common evaluation tasks.
 
 All evaluators accept the following common arguments:
-- model: Model to use (defaults to gpt-5-mini)
+- model: Model to use (defaults to gpt-5.6-terra)
 - temperature: Controls randomness (0-1). If not specified, uses the model's default.
 - max_tokens: Maximum tokens to generate. If not specified, uses the model's default.
 - client: OpenAI client (defaults to global client from init())

py/autoevals/oai.py

@@ -437,7 +437,7 @@ def init(
         default_model: The default model(s) to use for evaluations when not specified per-call.
             Can be either:
             - A string (for backward compatibility): Sets the default completion model only.
-              Defaults to "gpt-5-mini" if not set.
+              Defaults to "gpt-5.6-terra" if not set.
             - A dictionary with "completion" and/or "embedding" keys: Allows setting default
               models for different evaluation types. Only the specified models are updated;
               others remain unchanged.
@@ -494,8 +494,8 @@ def init(
 
 
 def get_default_model() -> str:
-    """Get the configured default completion model, or "gpt-5-mini" if not set."""
-    return _default_model_var.get(None) or "gpt-5-mini"
+    """Get the configured default completion model, or "gpt-5.6-terra" if not set."""
+    return _default_model_var.get(None) or "gpt-5.6-terra"
 
 
 def get_default_embedding_model() -> str:

py/autoevals/ragas.py

@@ -17,7 +17,7 @@
 
 **Common arguments**:
 
-    - `model`: Model to use for evaluation, defaults to the model configured via init(default_model=...) or "gpt-5-mini"
+    - `model`: Model to use for evaluation, defaults to the model configured via init(default_model=...) or "gpt-5.6-terra"
     - `client`: Optional Client for API calls. If not provided, uses global client from init()
 
 **Example - Direct usage**:
@@ -132,7 +132,7 @@ def check_required(name, **kwargs):
 
 # Deprecated: Use init(default_model="...") to configure the default model instead.
 # This was previously "gpt-5-nano" but now defaults to the configured model.
-DEFAULT_RAGAS_MODEL = "gpt-5-nano"
+DEFAULT_RAGAS_MODEL = "gpt-5.6-luna"
 
 
 def _get_model(model: str | None) -> str:
@@ -145,7 +145,7 @@ def _get_model(model: str | None) -> str:
         return model
 
     # Check if user configured a custom default via init(default_model=...)
-    # If they did (even if it's "gpt-5-mini"), respect it for consistency
+    # If they did (even if it's "gpt-5.6-terra"), respect it for consistency
     configured_default = _default_model_var.get(None)
     if configured_default is not None:
         return configured_default

py/autoevals/test_init_models.py

@@ -35,7 +35,7 @@ def test_object_form_can_set_embedding_model_only():
 
     assert get_default_embedding_model() == "text-embedding-3-large"
     # Completion model should remain at default since we didn't update it
-    assert get_default_model() == "gpt-5-mini"
+    assert get_default_model() == "gpt-5.6-terra"
 
 
 def test_object_form_can_set_both_models():
@@ -70,7 +70,7 @@ def test_falls_back_to_defaults_when_not_set():
     """Test that defaults are used when default_model is not provided."""
     init()
 
-    assert get_default_model() == "gpt-5-mini"
+    assert get_default_model() == "gpt-5.6-terra"
     assert get_default_embedding_model() == "text-embedding-ada-002"
 
 

py/autoevals/test_oai.py

@@ -262,11 +262,11 @@ def test_prepare_openai_v0_with_client(mock_openai_v0: OpenAIV0Module):
     assert prepared_client.complete.__name__ == "acreate"
 
 
-def test_get_default_model_returns_gpt_4o_by_default():
-    """Test that get_default_model returns gpt-5-mini when no default is configured."""
+def test_get_default_model_returns_gpt_5_6_terra_by_default():
+    """Test that get_default_model returns gpt-5.6-terra when no default is configured."""
     # Reset init to clear any previous default model
     init(None)
-    assert get_default_model() == "gpt-5-mini"
+    assert get_default_model() == "gpt-5.6-terra"
 
 
 def test_init_sets_default_model():
@@ -284,7 +284,7 @@ def test_init_can_reset_default_model():
     assert get_default_model() == "claude-3-5-sonnet-20241022"
 
     init(None, default_model=None)
-    assert get_default_model() == "gpt-5-mini"
+    assert get_default_model() == "gpt-5.6-terra"
 
 
 def test_init_can_set_both_client_and_default_model():