diff --git a/MIGRATION_GUIDE.md b/MIGRATION_GUIDE.md
index d0c895a878..a57452e67b 100644
--- a/MIGRATION_GUIDE.md
+++ b/MIGRATION_GUIDE.md
@@ -145,6 +145,7 @@ Looking to upgrade from Sentry SDK 2.x to 3.x? Here's a comprehensive list of wh
 - The `enable_logs` option was removed. Using Sentry's logging API now works without requiring setting `enable_logs=True`.
 - The `enable_metrics` option was removed.
 - The deprecated `@ai_track` decorator was removed.
+- The deprecated `record_token_usage` function was removed.
 - The deprecated `push_scope` and `configure_scope` APIs have been removed. Use `with new_scope():` to push a new scope and `scope = get_current_scope()` to retrieve the current scope instead.
 - The experimental `max_spans` option was removed.
 - The experimental `before_send_log` option was removed. Use the top-level `before_send_log` instead.
diff --git a/sentry_sdk/ai/monitoring.py b/sentry_sdk/ai/monitoring.py
deleted file mode 100644
index 4799e1892d..0000000000
--- a/sentry_sdk/ai/monitoring.py
+++ /dev/null
@@ -1,55 +0,0 @@
-from typing import TYPE_CHECKING
-
-from sentry_sdk.consts import SPANDATA
-from sentry_sdk.utils import deprecation_warning
-
-if TYPE_CHECKING:
-    from typing import Any, Awaitable, Callable, Optional, TypeVar, Union
-
-    from sentry_sdk.traces import Span
-
-    F = TypeVar("F", bound=Union[Callable[..., Any], Callable[..., Awaitable[Any]]])
-
-
-def record_token_usage(
-    span: "Span",
-    input_tokens: "Optional[int]" = None,
-    input_tokens_cached: "Optional[int]" = None,
-    input_tokens_cache_write: "Optional[int]" = None,
-    output_tokens: "Optional[int]" = None,
-    output_tokens_reasoning: "Optional[int]" = None,
-    total_tokens: "Optional[int]" = None,
-) -> None:
-    deprecation_warning(
-        "record_token_usage() is deprecated and will be removed in version 3.0 of sentry-sdk. "
-        "Use the manual span API instead, e.g. span.set_attribute() (in streaming mode) or span.set_data().",
-    )
-
-    if input_tokens is not None:
-        span.set_attribute(SPANDATA.GEN_AI_USAGE_INPUT_TOKENS, input_tokens)
-
-    if input_tokens_cached is not None:
-        span.set_attribute(
-            SPANDATA.GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS, input_tokens_cached
-        )
-
-    if input_tokens_cache_write is not None:
-        span.set_attribute(
-            SPANDATA.GEN_AI_USAGE_CACHE_CREATION_INPUT_TOKENS,
-            input_tokens_cache_write,
-        )
-
-    if output_tokens is not None:
-        span.set_attribute(SPANDATA.GEN_AI_USAGE_OUTPUT_TOKENS, output_tokens)
-
-    if output_tokens_reasoning is not None:
-        span.set_attribute(
-            SPANDATA.GEN_AI_USAGE_REASONING_OUTPUT_TOKENS,
-            output_tokens_reasoning,
-        )
-
-    if total_tokens is None and input_tokens is not None and output_tokens is not None:
-        total_tokens = input_tokens + output_tokens
-
-    if total_tokens is not None:
-        span.set_attribute(SPANDATA.GEN_AI_USAGE_TOTAL_TOKENS, total_tokens)