raylib 4.5
INFO: GL: OpenGL device information:
INFO: > Vendor: Intel
INFO: > Renderer: Mesa Intel(R) UHD Graphics (TGL GT1)
INFO: > Version: 4.6 (Core Profile) Mesa 21.2.6
INFO: > GLSL: 4.60

I think it could be related to the way the gltf is created. raylib uses cgltf library to load models.

I set the log level to debug and got this warning:

WARNING: MODEL: [asset/models/aircraft_carrier.glb] Indices data converted from u32 to u16, possible loss of data
WARNING: MODEL: [asset/models/aircraft_carrier.glb] Indices data converted from u32 to u16, possible loss of data
WARNING: MODEL: [asset/models/aircraft_carrier.glb] Indices data converted from u32 to u16, possible loss of data
WARNING: MODEL: [asset/models/aircraft_carrier.glb] Indices data converted from u32 to u16, possible loss of data

I also noticed that there is a restriction warning in rmodels.c:LoadGLTF,

- Vertex attibute types and formats supported:
    > Vertices (position): vec3: float
    > Normals: vec3: float
    > Texcoords: vec2: float
    > Colors: vec4: u8, u16, f32 (normalized)
    > Indices: u16, u32 (truncated to u16)

Does it mean raylib is only able to load mesh models with vertex number less than 65535 (max val of uint16_t) ?

I just ran into this last week and you're on the right track. I'll try to explain it to the extent that I understand it.

GLTF is an indexed model format, meaning it lists out all the vertex coordinates in the mesh, and then uses sets of 3 indexes (like vertex 0 2 3, 2 3 5, etc) to make each triangle. For large models like yours, it is more efficient with space, but it takes more processing power for the GPU to compute and has that pesky vertex count limit which I believe comes from the functionality of the cgltf library that @raysan5 mentioned.

To solve this, you can use a non-indexed model format like .obj, which instead lists the position of each vertex in each triangle for every triangle. This makes the file format larger, but it is easier for the GPU to run through and draw the model. And it won't have the same indexing limit that the GLTF loader has.

Great thanks for your detailed reply ! That helps a lot for me to understand the current design.
I just tried the obj format, but it seems the vertex/face number could still be a problem.

test_model_loading: /path/to/xxx-project/build/_deps/raylib-src/src/external/tinyobj_loader_c.h:1105: parseLine: Assertion `3 * num_f < TINYOBJ_MAX_FACES_PER_F_LINE' failed.

I guess it is necessary to compress my model (vertex/edge/face numbers) in blender before use it in raylib.

This issue seems to be external to raylib, maybe related to cgltf.h library or how the model is exported from the 3d creation software. Please note that glTF format is a complex format with many export options and also several versions, not all possibilities are supported.