diff --git a/crates/ty_ide/src/completion.rs b/crates/ty_ide/src/completion.rs
index 2470308c41..dc7072dfd7 100644
--- a/crates/ty_ide/src/completion.rs
+++ b/crates/ty_ide/src/completion.rs
@@ -23,7 +23,18 @@ pub fn completion(db: &dyn Db, file: File, offset: TextSize) -> Vec<Completion<'
     let model = SemanticModel::new(db, file);
     let mut completions = match target {
         CompletionTargetAst::ObjectDot { expr } => model.attribute_completions(expr),
-        CompletionTargetAst::ImportFrom { import, name } => model.import_completions(import, name),
+        CompletionTargetAst::ObjectDotInImport { import, name } => {
+            model.import_submodule_completions(import, name)
+        }
+        CompletionTargetAst::ObjectDotInImportFrom { import } => {
+            model.from_import_submodule_completions(import)
+        }
+        CompletionTargetAst::ImportFrom { import, name } => {
+            model.from_import_completions(import, name)
+        }
+        CompletionTargetAst::Import { .. } | CompletionTargetAst::ImportViaFrom { .. } => {
+            model.import_completions()
+        }
         CompletionTargetAst::Scoped { node } => model.scoped_completions(node),
     };
     completions.sort_by(compare_suggestions);
@@ -50,11 +61,11 @@ enum CompletionTargetTokens<'t> {
         object: &'t Token,
         /// The token, if non-empty, following the dot.
         ///
-        /// This is currently unused, but we should use this
-        /// eventually to remove completions that aren't a
-        /// prefix of what has already been typed. (We are
-        /// currently relying on the LSP client to do this.)
-        #[expect(dead_code)]
+        /// For right now, this is only used to determine which
+        /// module in an `import` statement to return submodule
+        /// completions for. But we could use it for other things,
+        /// like only returning completions that start with a prefix
+        /// corresponding to this token.
         attribute: Option<&'t Token>,
     },
     /// A `from module import attribute` token form was found, where
@@ -63,6 +74,20 @@ enum CompletionTargetTokens<'t> {
         /// The module being imported from.
         module: &'t Token,
     },
+    /// A `import module` token form was found, where `module` may be
+    /// empty.
+    Import {
+        /// The token corresponding to the `import` keyword.
+        import: &'t Token,
+        /// The token closest to the cursor.
+        ///
+        /// This is currently unused, but we should use this
+        /// eventually to remove completions that aren't a
+        /// prefix of what has already been typed. (We are
+        /// currently relying on the LSP client to do this.)
+        #[expect(dead_code)]
+        module: &'t Token,
+    },
     /// A token was found under the cursor, but it didn't
     /// match any of our anticipated token patterns.
     Generic { token: &'t Token },
@@ -105,6 +130,8 @@ impl<'t> CompletionTargetTokens<'t> {
                 }
             } else if let Some(module) = import_from_tokens(before) {
                 CompletionTargetTokens::ImportFrom { module }
+            } else if let Some((import, module)) = import_tokens(before) {
+                CompletionTargetTokens::Import { import, module }
             } else if let Some([_]) = token_suffix_by_kinds(before, [TokenKind::Float]) {
                 // If we're writing a `float`, then we should
                 // specifically not offer completions. This wouldn't
@@ -140,19 +167,47 @@ impl<'t> CompletionTargetTokens<'t> {
         offset: TextSize,
     ) -> Option<CompletionTargetAst<'t>> {
         match *self {
-            CompletionTargetTokens::PossibleObjectDot { object, .. } => {
+            CompletionTargetTokens::PossibleObjectDot { object, attribute } => {
                 let covering_node = covering_node(parsed.syntax().into(), object.range())
-                    // We require that the end of the node range not
-                    // exceed the cursor offset. This avoids selecting
-                    // a node "too high" in the AST in cases where
-                    // completions are requested in the middle of an
-                    // expression. e.g., `foo.<CURSOR>.bar`.
-                    .find_last(|node| node.is_expr_attribute() && node.range().end() <= offset)
+                    .find_last(|node| {
+                        // We require that the end of the node range not
+                        // exceed the cursor offset. This avoids selecting
+                        // a node "too high" in the AST in cases where
+                        // completions are requested in the middle of an
+                        // expression. e.g., `foo.<CURSOR>.bar`.
+                        if node.is_expr_attribute() {
+                            return node.range().end() <= offset;
+                        }
+                        // For import statements though, they can't be
+                        // nested, so we don't care as much about the
+                        // cursor being strictly after the statement.
+                        // And indeed, sometimes it won't be! e.g.,
+                        //
+                        //   import re, os.p<CURSOR>, zlib
+                        //
+                        // So just return once we find an import.
+                        node.is_stmt_import() || node.is_stmt_import_from()
+                    })
                     .ok()?;
                 match covering_node.node() {
                     ast::AnyNodeRef::ExprAttribute(expr) => {
                         Some(CompletionTargetAst::ObjectDot { expr })
                     }
+                    ast::AnyNodeRef::StmtImport(import) => {
+                        let range = attribute
+                            .map(Ranged::range)
+                            .unwrap_or_else(|| object.range());
+                        // Find the name that overlaps with the
+                        // token we identified for the attribute.
+                        let name = import
+                            .names
+                            .iter()
+                            .position(|alias| alias.range().contains_range(range))?;
+                        Some(CompletionTargetAst::ObjectDotInImport { import, name })
+                    }
+                    ast::AnyNodeRef::StmtImportFrom(import) => {
+                        Some(CompletionTargetAst::ObjectDotInImportFrom { import })
+                    }
                     _ => None,
                 }
             }
@@ -165,6 +220,20 @@ impl<'t> CompletionTargetTokens<'t> {
                 };
                 Some(CompletionTargetAst::ImportFrom { import, name: None })
             }
+            CompletionTargetTokens::Import { import, .. } => {
+                let covering_node = covering_node(parsed.syntax().into(), import.range())
+                    .find_first(|node| node.is_stmt_import() || node.is_stmt_import_from())
+                    .ok()?;
+                match covering_node.node() {
+                    ast::AnyNodeRef::StmtImport(import) => {
+                        Some(CompletionTargetAst::Import { import, name: None })
+                    }
+                    ast::AnyNodeRef::StmtImportFrom(import) => {
+                        Some(CompletionTargetAst::ImportViaFrom { import })
+                    }
+                    _ => None,
+                }
+            }
             CompletionTargetTokens::Generic { token } => {
                 let covering_node = covering_node(parsed.syntax().into(), token.range());
                 Some(CompletionTargetAst::Scoped {
@@ -188,6 +257,18 @@ enum CompletionTargetAst<'t> {
     /// A `object.attribute` scenario, where we want to
     /// list attributes on `object` for completions.
     ObjectDot { expr: &'t ast::ExprAttribute },
+    /// A `import module.submodule` scenario, where we only want to
+    /// list submodules for completions.
+    ObjectDotInImport {
+        /// The import statement.
+        import: &'t ast::StmtImport,
+        /// An index into `import.names`. The index is guaranteed to be
+        /// valid.
+        name: usize,
+    },
+    /// A `from module.submodule` scenario, where we only want to list
+    /// submodules for completions.
+    ObjectDotInImportFrom { import: &'t ast::StmtImportFrom },
     /// A `from module import attribute` scenario, where we want to
     /// list attributes on `module` for completions.
     ImportFrom {
@@ -197,6 +278,24 @@ enum CompletionTargetAst<'t> {
         /// set, the index is guaranteed to be valid.
         name: Option<usize>,
     },
+    /// A `import module` scenario, where we want to
+    /// list available modules for completions.
+    Import {
+        /// The import statement.
+        #[expect(dead_code)]
+        import: &'t ast::StmtImport,
+        /// An index into `import.names` if relevant. When this is
+        /// set, the index is guaranteed to be valid.
+        #[expect(dead_code)]
+        name: Option<usize>,
+    },
+    /// A `from module` scenario, where we want to
+    /// list available modules for completions.
+    ImportViaFrom {
+        /// The import statement.
+        #[expect(dead_code)]
+        import: &'t ast::StmtImportFrom,
+    },
     /// A scoped scenario, where we want to list all items available in
     /// the most narrow scope containing the giving AST node.
     Scoped { node: ast::AnyNodeRef<'t> },
@@ -317,6 +416,52 @@ fn import_from_tokens(tokens: &[Token]) -> Option<&Token> {
     None
 }
 
+/// Looks for the start of a `import <CURSOR>` statement.
+///
+/// This also handles cases like `import foo, c<CURSOR>, bar`.
+///
+/// If found, a token corresponding to the `import` or `from` keyword
+/// and the the closest point of the `<CURSOR>` is returned.
+///
+/// It is assumed that callers will call `from_import_tokens` first to
+/// try and recognize a `from ... import ...` statement before using
+/// this.
+fn import_tokens(tokens: &[Token]) -> Option<(&Token, &Token)> {
+    use TokenKind as TK;
+
+    /// A look-back limit, in order to bound work.
+    ///
+    /// See `LIMIT` in `import_from_tokens` for more context.
+    const LIMIT: usize = 1_000;
+
+    /// A state used to "parse" the tokens preceding the user's cursor,
+    /// in reverse, to detect a `import` statement.
+    enum S {
+        Start,
+        Names,
+    }
+
+    let mut state = S::Start;
+    let module_token = tokens.last()?;
+    // Move backward through the tokens until we get to
+    // the `import` token.
+    for token in tokens.iter().rev().take(LIMIT) {
+        state = match (state, token.kind()) {
+            // It's okay to pop off a newline token here initially,
+            // since it may occur when the name being imported is
+            // empty.
+            (S::Start, TK::Newline) => S::Names,
+            // Munch through tokens that can make up an alias.
+            (S::Start | S::Names, TK::Name | TK::Comma | TK::As | TK::Unknown) => S::Names,
+            (S::Start | S::Names, TK::Import | TK::From) => {
+                return Some((token, module_token));
+            }
+            _ => return None,
+        };
+    }
+    None
+}
+
 /// Order completions lexicographically, with these exceptions:
 ///
 /// 1) A `_[^_]` prefix sorts last and
@@ -2709,6 +2854,143 @@ importlib.<CURSOR>
         test.assert_completions_include("resources");
     }
 
+    #[test]
+    fn import_with_leading_character() {
+        let test = cursor_test(
+            "\
+import c<CURSOR>
+",
+        );
+        test.assert_completions_include("collections");
+    }
+
+    #[test]
+    fn import_without_leading_character() {
+        let test = cursor_test(
+            "\
+import <CURSOR>
+",
+        );
+        test.assert_completions_include("collections");
+    }
+
+    #[test]
+    fn import_multiple() {
+        let test = cursor_test(
+            "\
+import re, c<CURSOR>, sys
+",
+        );
+        test.assert_completions_include("collections");
+    }
+
+    #[test]
+    fn import_with_aliases() {
+        let test = cursor_test(
+            "\
+import re as regexp, c<CURSOR>, sys as system
+",
+        );
+        test.assert_completions_include("collections");
+    }
+
+    #[test]
+    fn import_over_multiple_lines() {
+        let test = cursor_test(
+            "\
+import re as regexp, \\
+    c<CURSOR>, \\
+    sys as system
+",
+        );
+        test.assert_completions_include("collections");
+    }
+
+    #[test]
+    fn import_unknown_in_module() {
+        let test = cursor_test(
+            "\
+import ?, <CURSOR>
+",
+        );
+        test.assert_completions_include("collections");
+    }
+
+    #[test]
+    fn import_via_from_with_leading_character() {
+        let test = cursor_test(
+            "\
+from c<CURSOR>
+",
+        );
+        test.assert_completions_include("collections");
+    }
+
+    #[test]
+    fn import_via_from_without_leading_character() {
+        let test = cursor_test(
+            "\
+from <CURSOR>
+",
+        );
+        test.assert_completions_include("collections");
+    }
+
+    #[test]
+    fn import_statement_with_submodule_with_leading_character() {
+        let test = cursor_test(
+            "\
+import os.p<CURSOR>
+",
+        );
+        test.assert_completions_include("path");
+        test.assert_completions_do_not_include("abspath");
+    }
+
+    #[test]
+    fn import_statement_with_submodule_multiple() {
+        let test = cursor_test(
+            "\
+import re, os.p<CURSOR>, zlib
+",
+        );
+        test.assert_completions_include("path");
+        test.assert_completions_do_not_include("abspath");
+    }
+
+    #[test]
+    fn import_statement_with_submodule_without_leading_character() {
+        let test = cursor_test(
+            "\
+import os.<CURSOR>
+",
+        );
+        test.assert_completions_include("path");
+        test.assert_completions_do_not_include("abspath");
+    }
+
+    #[test]
+    fn import_via_from_with_submodule_with_leading_character() {
+        let test = cursor_test(
+            "\
+from os.p<CURSOR>
+",
+        );
+        test.assert_completions_include("path");
+        test.assert_completions_do_not_include("abspath");
+    }
+
+    #[test]
+    fn import_via_from_with_submodule_without_leading_character() {
+        let test = cursor_test(
+            "\
+from os.<CURSOR>
+",
+        );
+        test.assert_completions_include("path");
+        test.assert_completions_do_not_include("abspath");
+    }
+
     #[test]
     fn regression_test_issue_642() {
         // Regression test for https://github.com/astral-sh/ty/issues/642
diff --git a/crates/ty_python_semantic/src/semantic_model.rs b/crates/ty_python_semantic/src/semantic_model.rs
index f038111c5f..47eacbb3a5 100644
--- a/crates/ty_python_semantic/src/semantic_model.rs
+++ b/crates/ty_python_semantic/src/semantic_model.rs
@@ -6,7 +6,7 @@ use ruff_source_file::LineIndex;
 
 use crate::Db;
 use crate::module_name::ModuleName;
-use crate::module_resolver::{KnownModule, Module, resolve_module};
+use crate::module_resolver::{KnownModule, Module, list_modules, resolve_module};
 use crate::semantic_index::definition::Definition;
 use crate::semantic_index::scope::FileScopeId;
 use crate::semantic_index::semantic_index;
@@ -41,8 +41,24 @@ impl<'db> SemanticModel<'db> {
         resolve_module(self.db, module_name)
     }
 
+    /// Returns completions for symbols available in a `import <CURSOR>` context.
+    pub fn import_completions(&self) -> Vec<Completion<'db>> {
+        list_modules(self.db)
+            .into_iter()
+            .map(|module| {
+                let builtin = module.is_known(self.db, KnownModule::Builtins);
+                let ty = Type::module_literal(self.db, self.file, module);
+                Completion {
+                    name: Name::new(module.name(self.db).as_str()),
+                    ty,
+                    builtin,
+                }
+            })
+            .collect()
+    }
+
     /// Returns completions for symbols available in a `from module import <CURSOR>` context.
-    pub fn import_completions(
+    pub fn from_import_completions(
         &self,
         import: &ast::StmtImportFrom,
         _name: Option<usize>,
@@ -61,6 +77,79 @@ impl<'db> SemanticModel<'db> {
         self.module_completions(&module_name)
     }
 
+    /// Returns completions only for submodules for the module
+    /// identified by `name` in `import`.
+    ///
+    /// For example, `import re, os.<CURSOR>, zlib`.
+    pub fn import_submodule_completions(
+        &self,
+        import: &ast::StmtImport,
+        name: usize,
+    ) -> Vec<Completion<'db>> {
+        let module_ident = &import.names[name].name;
+        let Some((parent_ident, _)) = module_ident.rsplit_once('.') else {
+            return vec![];
+        };
+        let module_name =
+            match ModuleName::from_identifier_parts(self.db, self.file, Some(parent_ident), 0) {
+                Ok(module_name) => module_name,
+                Err(err) => {
+                    tracing::debug!(
+                        "Could not extract module name from `{module:?}`: {err:?}",
+                        module = module_ident,
+                    );
+                    return vec![];
+                }
+            };
+        self.import_submodule_completions_for_name(&module_name)
+    }
+
+    /// Returns completions only for submodules for the module
+    /// used in a `from module import attribute` statement.
+    ///
+    /// For example, `from os.<CURSOR>`.
+    pub fn from_import_submodule_completions(
+        &self,
+        import: &ast::StmtImportFrom,
+    ) -> Vec<Completion<'db>> {
+        let level = import.level;
+        let Some(module_ident) = import.module.as_deref() else {
+            return vec![];
+        };
+        let Some((parent_ident, _)) = module_ident.rsplit_once('.') else {
+            return vec![];
+        };
+        let module_name = match ModuleName::from_identifier_parts(
+            self.db,
+            self.file,
+            Some(parent_ident),
+            level,
+        ) {
+            Ok(module_name) => module_name,
+            Err(err) => {
+                tracing::debug!(
+                    "Could not extract module name from `{module:?}` with level {level}: {err:?}",
+                    module = import.module,
+                    level = import.level,
+                );
+                return vec![];
+            }
+        };
+        self.import_submodule_completions_for_name(&module_name)
+    }
+
+    /// Returns submodule-only completions for the given module.
+    fn import_submodule_completions_for_name(
+        &self,
+        module_name: &ModuleName,
+    ) -> Vec<Completion<'db>> {
+        let Some(module) = resolve_module(self.db, module_name) else {
+            tracing::debug!("Could not resolve module from `{module_name:?}`");
+            return vec![];
+        };
+        self.submodule_completions(&module)
+    }
+
     /// Returns completions for symbols available in the given module as if
     /// it were imported by this model's `File`.
     fn module_completions(&self, module_name: &ModuleName) -> Vec<Completion<'db>> {
@@ -75,11 +164,20 @@ impl<'db> SemanticModel<'db> {
         for crate::types::Member { name, ty } in crate::types::all_members(self.db, ty) {
             completions.push(Completion { name, ty, builtin });
         }
+        completions.extend(self.submodule_completions(&module));
+        completions
+    }
+
+    /// Returns completions for submodules of the given module.
+    fn submodule_completions(&self, module: &Module<'db>) -> Vec<Completion<'db>> {
+        let builtin = module.is_known(self.db, KnownModule::Builtins);
+
+        let mut completions = vec![];
         for submodule_basename in module.all_submodules(self.db) {
             let Some(basename) = ModuleName::new(submodule_basename.as_str()) else {
                 continue;
             };
-            let mut submodule_name = module_name.clone();
+            let mut submodule_name = module.name(self.db).clone();
             submodule_name.extend(&basename);
 
             let Some(submodule) = resolve_module(self.db, &submodule_name) else {