diff --git a/src/allotropy/allotrope/schema_parser/generate_schemas.py b/src/allotropy/allotrope/schema_parser/generate_schemas.py
index 235611039..a1c0a052c 100644
--- a/src/allotropy/allotrope/schema_parser/generate_schemas.py
+++ b/src/allotropy/allotrope/schema_parser/generate_schemas.py
@@ -32,6 +32,16 @@
 from allotropy.allotrope.schema_parser.update_units import update_unit_files
 from allotropy.allotrope.schemas import get_schema
 
+# Optimized version for better performance
+try:
+    from allotropy.allotrope.schema_parser.generate_schemas_optimized import (
+        generate_schemas_optimized,
+    )
+
+    HAS_OPTIMIZED_VERSION = True
+except ImportError:
+    HAS_OPTIMIZED_VERSION = False
+
 
 def lint_file(model_path: Path) -> None:
     # The first run of ruff changes typing annotations and causes unused imports. We catch failure
@@ -143,12 +153,20 @@ def generate_schemas(
     *,
     dry_run: bool | None = False,
     schema_regex: str | None = None,
+    use_optimized: bool = True,
 ) -> list[str]:
     """Generate schemas from JSON schema files.
     :dry_run: If true, does not save changes to any models, but still returns the list of models that would change.
     :schema_regex: If set, filters schemas to generate using regex.
+    :use_optimized: If true and available, use the optimized parallel version
     :return: A list of model files that were changed.
     """
+    # Use optimized version if available and requested
+    if use_optimized and HAS_OPTIMIZED_VERSION:
+        return generate_schemas_optimized(
+            dry_run=bool(dry_run),
+            schema_regex=schema_regex,
+        )
     unit_to_iri: dict[str, str] = {}
     schema_paths = list(Path(SCHEMA_DIR_PATH).rglob("*.json"))
     models_changed = []
diff --git a/src/allotropy/allotrope/schema_parser/generate_schemas_optimized.py b/src/allotropy/allotrope/schema_parser/generate_schemas_optimized.py
new file mode 100644
index 000000000..81f4565ca
--- /dev/null
+++ b/src/allotropy/allotrope/schema_parser/generate_schemas_optimized.py
@@ -0,0 +1,225 @@
+"""Optimized schema generation that runs in memory and in parallel."""
+
+from concurrent.futures import as_completed, ProcessPoolExecutor
+import json
+from pathlib import Path
+import re
+import tempfile
+from typing import Any
+import warnings
+
+from autoflake import fix_code  # type: ignore[import-untyped]
+import black
+from datamodel_code_generator import (
+    DataModelType,
+    generate,
+    InputFileType,
+    PythonVersion,
+)
+
+from allotropy.allotrope.schema_parser.backup_manager import (
+    backup_paths,
+    get_original_path,
+)
+from allotropy.allotrope.schema_parser.model_class_editor import (
+    get_shared_schema_info,
+    ModelClassEditor,
+)
+from allotropy.allotrope.schema_parser.path_util import (
+    GENERATED_SHARED_PATHS,
+    get_manifest_from_schema_path,
+    get_model_path_from_schema_path,
+    get_rel_schema_path,
+    get_schema_path_from_model_path,
+    MODEL_DIR_PATH,
+    SCHEMA_DIR_PATH,
+)
+from allotropy.allotrope.schema_parser.schema_cleaner import SchemaCleaner
+from allotropy.allotrope.schema_parser.update_units import update_unit_files
+from allotropy.allotrope.schemas import get_schema
+
+
+def format_code_in_memory(code: str) -> str:
+    """Format Python code in memory using black and autoflake."""
+    # First pass: fix imports using autoflake
+    code = fix_code(
+        code,
+        remove_all_unused_imports=True,
+        remove_unused_variables=True,
+        remove_duplicate_keys=True,
+        ignore_init_module_imports=False,
+    )
+
+    # Second pass: format with black
+    try:
+        code = black.format_str(code, mode=black.Mode())  # type: ignore[attr-defined]
+    except Exception:  # noqa: S110
+        # If black fails, return the autoflake-cleaned code
+        pass
+
+    return code
+
+
+def generate_model_in_memory(schema: dict[str, Any]) -> str:
+    """Generate model code from schema without writing to disk."""
+    with warnings.catch_warnings():
+        # Suppress expected warnings
+        warnings.filterwarnings(
+            "ignore",
+            category=UserWarning,
+            message="Field name `.*` is duplicated on .*",
+        )
+        warnings.filterwarnings(
+            "ignore",
+            category=UserWarning,
+            message=re.escape(
+                "format of 'iri' not understood for 'string' - using default"
+            ),
+        )
+
+        # Use a temporary file for the schema (datamodel_code_generator requires a file)
+        with tempfile.NamedTemporaryFile(mode="w", suffix=".json") as schema_file:
+            json.dump(schema, schema_file)
+            schema_file.flush()
+
+            # Generate to a temporary output file
+            with tempfile.NamedTemporaryFile(mode="w+", suffix=".py") as output_file:
+                generate(
+                    input_=Path(schema_file.name),
+                    output=Path(output_file.name),
+                    output_model_type=DataModelType.DataclassesDataclass,
+                    input_file_type=InputFileType.JsonSchema,
+                    base_class="",
+                    target_python_version=PythonVersion.PY_310,
+                    use_union_operator=True,
+                )
+
+                # Read the generated content
+                output_file.seek(0)
+                generated_code = output_file.read()
+
+    return generated_code
+
+
+def process_single_schema(
+    schema_path: Path, schema_regex: str | None = None
+) -> tuple[str | None, dict[str, str], str | None]:
+    """Process a single schema file and return (model_name, units, generated_code)."""
+    rel_schema_path = str(get_rel_schema_path(schema_path))
+
+    # Check if should process
+    if rel_schema_path.startswith("shared"):
+        return None, {}, None
+    if schema_regex and not re.match(schema_regex, rel_schema_path):
+        return None, {}, None
+
+    print(f"Generating models for schema: {rel_schema_path}...")  # noqa: T201
+
+    # Clean schema and get units
+    schema_cleaner = SchemaCleaner()
+    schema = schema_cleaner.clean(get_schema(schema_path))
+    unit_to_iri = schema_cleaner.get_referenced_units()
+
+    # Generate model code in memory
+    generated_code = generate_model_in_memory(schema)
+
+    # Modify the generated code (imports, etc.)
+    model_path = Path(MODEL_DIR_PATH, get_model_path_from_schema_path(schema_path))
+    schema_path_for_manifest = get_schema_path_from_model_path(
+        get_original_path(model_path)
+    )
+
+    classes_to_skip, imports_to_add = get_shared_schema_info(schema)
+    manifest = get_manifest_from_schema_path(schema_path_for_manifest)
+    editor = ModelClassEditor(
+        manifest, classes_to_skip, imports_to_add, schema_path_for_manifest.name
+    )
+    modified_code = editor.modify_file(generated_code)
+
+    # Format the code in memory
+    formatted_code = format_code_in_memory(modified_code)
+
+    model_path = Path(MODEL_DIR_PATH, get_model_path_from_schema_path(schema_path))
+    return model_path.stem, unit_to_iri, formatted_code
+
+
+def generate_schemas_optimized(
+    *,
+    dry_run: bool = False,
+    schema_regex: str | None = None,
+    max_workers: int | None = None,
+) -> list[str]:
+    """
+    Generate schemas with in-memory processing and parallelization.
+
+    This version:
+    - Processes schemas in parallel using multiprocessing
+    - Generates and formats code in memory
+    - Only writes final results (no temp files)
+    - Reduces I/O operations significantly
+    :param dry_run: If true, does not save changes to any models
+    :param schema_regex: If set, filters schemas to generate using regex
+    :param max_workers: Maximum number of parallel workers (default: CPU count)
+    :return: List of model files that were changed
+    """
+    schema_paths = list(Path(SCHEMA_DIR_PATH).rglob("*.json"))
+
+    # Process schemas in parallel
+    models_changed = []
+    all_units = {}
+    generated_models = {}
+
+    with ProcessPoolExecutor(max_workers=max_workers) as executor:
+        # Submit all tasks
+        future_to_path = {
+            executor.submit(process_single_schema, path, schema_regex): path
+            for path in schema_paths
+        }
+
+        # Collect results as they complete
+        for future in as_completed(future_to_path):
+            schema_path = future_to_path[future]
+            try:
+                model_name, units, code = future.result()
+                if model_name:
+                    all_units.update(units)
+                    model_path = Path(
+                        MODEL_DIR_PATH, get_model_path_from_schema_path(schema_path)
+                    )
+                    generated_models[model_path] = code
+            except Exception as e:
+                print(f"Error processing {schema_path}: {e}")  # noqa: T201
+
+    # Now write all results (unless dry_run)
+    for model_path, code in generated_models.items():
+        # Ensure directory exists
+        model_path.parent.mkdir(parents=True, exist_ok=True)
+
+        # Check if file changed
+        if model_path.exists():
+            with open(model_path) as f:
+                existing_code = f.read()
+            if existing_code == code:
+                continue  # No change
+
+        # Write the new code
+        if not dry_run and code is not None:
+            with open(model_path, "w") as f:
+                f.write(code)
+
+        models_changed.append(model_path.stem)
+
+    # Update unit files (this must be done serially)
+    if not dry_run:
+        with backup_paths(GENERATED_SHARED_PATHS, restore=False) as working_paths:
+            update_unit_files(all_units, *working_paths)
+            # Format the unit files
+            for path in working_paths:
+                if path.suffix == ".py" and path.exists():
+                    with open(path) as f:
+                        content = f.read()
+                    formatted = format_code_in_memory(content)
+                    with open(path, "w") as f:
+                        f.write(formatted)
+
+    return models_changed