diff --git a/cpp/functions/gfx/mesh_gpu.cpp b/cpp/functions/gfx/mesh_gpu.cpp
new file mode 100644
index 00000000..6b8ea084
--- /dev/null
+++ b/cpp/functions/gfx/mesh_gpu.cpp
@@ -0,0 +1,69 @@
+#include "gfx/mesh_gpu.h"
+
+#include <vector>
+
+namespace fn::gfx {
+
+MeshGpu mesh_gpu_upload(const Mesh& mesh) {
+    MeshGpu g{};
+    if (mesh.positions.empty() || mesh.indices.empty()) return g;
+    if (mesh.normals.size() != mesh.positions.size()) return g;
+
+    const size_t nverts = mesh.positions.size() / 3;
+
+    // Interleave pos.xyz + normal.xyz, stride 6 floats.
+    std::vector<float> interleaved;
+    interleaved.resize(nverts * 6);
+    for (size_t i = 0; i < nverts; ++i) {
+        interleaved[i*6 + 0] = mesh.positions[i*3 + 0];
+        interleaved[i*6 + 1] = mesh.positions[i*3 + 1];
+        interleaved[i*6 + 2] = mesh.positions[i*3 + 2];
+        interleaved[i*6 + 3] = mesh.normals[i*3 + 0];
+        interleaved[i*6 + 4] = mesh.normals[i*3 + 1];
+        interleaved[i*6 + 5] = mesh.normals[i*3 + 2];
+    }
+
+    glGenVertexArrays(1, &g.vao);
+    glGenBuffers(1, &g.vbo);
+    glGenBuffers(1, &g.ebo);
+
+    glBindVertexArray(g.vao);
+
+    glBindBuffer(GL_ARRAY_BUFFER, g.vbo);
+    glBufferData(GL_ARRAY_BUFFER,
+                 (GLsizeiptr)(interleaved.size() * sizeof(float)),
+                 interleaved.data(),
+                 GL_STATIC_DRAW);
+
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, g.ebo);
+    glBufferData(GL_ELEMENT_ARRAY_BUFFER,
+                 (GLsizeiptr)(mesh.indices.size() * sizeof(uint32_t)),
+                 mesh.indices.data(),
+                 GL_STATIC_DRAW);
+
+    // location 0 = a_pos (vec3)
+    glEnableVertexAttribArray(0);
+    glVertexAttribPointer(0, 3, GL_FLOAT, GL_FALSE,
+                          (GLsizei)(6 * sizeof(float)),
+                          (const void*)0);
+    // location 1 = a_normal (vec3)
+    glEnableVertexAttribArray(1);
+    glVertexAttribPointer(1, 3, GL_FLOAT, GL_FALSE,
+                          (GLsizei)(6 * sizeof(float)),
+                          (const void*)(3 * sizeof(float)));
+
+    glBindBuffer(GL_ARRAY_BUFFER, 0);
+    glBindVertexArray(0);
+
+    g.index_count = (int)mesh.indices.size();
+    return g;
+}
+
+void mesh_gpu_destroy(MeshGpu& g) {
+    if (g.ebo) { glDeleteBuffers(1, &g.ebo); g.ebo = 0; }
+    if (g.vbo) { glDeleteBuffers(1, &g.vbo); g.vbo = 0; }
+    if (g.vao) { glDeleteVertexArrays(1, &g.vao); g.vao = 0; }
+    g.index_count = 0;
+}
+
+} // namespace fn::gfx
diff --git a/cpp/functions/gfx/mesh_gpu.h b/cpp/functions/gfx/mesh_gpu.h
new file mode 100644
index 00000000..e1f37576
--- /dev/null
+++ b/cpp/functions/gfx/mesh_gpu.h
@@ -0,0 +1,25 @@
+#pragma once
+
+#include "gfx/gl_loader.h"
+#include "gfx/mesh_obj_load.h"
+
+namespace fn::gfx {
+
+// VAO + VBO interleaved (pos.xyz, nrm.xyz) + EBO. ok() despues de upload.
+struct MeshGpu {
+    GLuint vao         = 0;
+    GLuint vbo         = 0;
+    GLuint ebo         = 0;
+    int    index_count = 0;
+    bool ok() const { return vao != 0 && index_count > 0; }
+};
+
+// Sube un Mesh CPU al GPU. Requiere contexto GL activo.
+// Si Mesh esta vacio o no es valido, devuelve MeshGpu{} (ok() == false).
+// Layout: location 0 = vec3 a_pos, location 1 = vec3 a_normal.
+MeshGpu mesh_gpu_upload(const Mesh& mesh);
+
+// Libera los recursos GL. Seguro con mesh_gpu vacio.
+void mesh_gpu_destroy(MeshGpu& mesh_gpu);
+
+} // namespace fn::gfx
diff --git a/cpp/functions/gfx/mesh_gpu.md b/cpp/functions/gfx/mesh_gpu.md
new file mode 100644
index 00000000..c311ce50
--- /dev/null
+++ b/cpp/functions/gfx/mesh_gpu.md
@@ -0,0 +1,68 @@
+---
+name: mesh_gpu
+kind: function
+lang: cpp
+domain: gfx
+version: "1.0.0"
+purity: impure
+signature: "MeshGpu mesh_gpu_upload(const Mesh&); void mesh_gpu_destroy(MeshGpu&)"
+description: "Sube un Mesh CPU a OpenGL como VAO + VBO interleaved (pos.xyz, normal.xyz) + EBO uint32. Layout: location 0 = a_pos vec3, location 1 = a_normal vec3, stride 6 floats."
+tags: [opengl, mesh, vao, vbo, ebo, gpu, gfx]
+uses_functions: [mesh_obj_load_cpp_gfx]
+uses_types: []
+returns: []
+returns_optional: false
+error_type: "error_go_core"
+imports: [GL/gl.h, GL/glext.h]
+tested: false
+tests: []
+test_file_path: ""
+file_path: "cpp/functions/gfx/mesh_gpu.cpp"
+framework: opengl
+params:
+  - name: mesh
+    desc: "Mesh CPU con positions/normals (mismo length, stride 3) e indices uint32. Si esta vacio o invalido, upload devuelve MeshGpu{} (ok()==false)."
+  - name: mesh_gpu
+    desc: "MeshGpu (vao/vbo/ebo, index_count). destroy libera todo y pone IDs a 0."
+output: "mesh_gpu_upload: MeshGpu listo para draw con glDrawElements(GL_TRIANGLES, index_count, GL_UNSIGNED_INT, 0). Si !ok(), no hubo upload."
+---
+
+# mesh_gpu
+
+CRUD GPU minimal para `Mesh`. Asume contexto OpenGL 3.3+ activo.
+
+## Layout de attribs
+
+```glsl
+#version 330 core
+layout(location = 0) in vec3 a_pos;
+layout(location = 1) in vec3 a_normal;
+```
+
+Stride = `6 * sizeof(float)`, sin padding.
+
+## Uso tipico
+
+```cpp
+auto cpu = fn::gfx::mesh_obj_load("model.obj");
+auto gpu = fn::gfx::mesh_gpu_upload(cpu);
+if (!gpu.ok()) { /* falla */ return; }
+
+// Draw:
+glUseProgram(prog);
+glBindVertexArray(gpu.vao);
+glDrawElements(GL_TRIANGLES, gpu.index_count, GL_UNSIGNED_INT, 0);
+glBindVertexArray(0);
+
+// Cleanup:
+fn::gfx::mesh_gpu_destroy(gpu);
+```
+
+## Validacion de input
+
+`mesh_gpu_upload` exige que `mesh.normals.size() == mesh.positions.size()`. `mesh_obj_parse` siempre genera normales (per-face si faltan) → invariante natural.
+
+## Notas
+
+- Indices son `GL_UNSIGNED_INT` (32-bit) para soportar meshes grandes sin tener que decidir formato dinamicamente.
+- `GL_STATIC_DRAW`: el assumption es que la malla no cambia post-upload. Si necesitas streaming, crear otro helper con `GL_DYNAMIC_DRAW`.