diff --git a/tools/server/server-models.cpp b/tools/server/server-models.cpp
index 35d21c8767..454522f04f 100644
--- a/tools/server/server-models.cpp
+++ b/tools/server/server-models.cpp
@@ -585,6 +585,11 @@ void server_model_meta::update_caps(const common_params & base) {
 
     // offline discovery cannot see video; a loaded model reports it
     architecture = server_model_architecture_json(inp_image, inp_audio, false, output_modalities);
+
+    // the trained context, read from the GGUF metadata
+    if (!params.model.path.empty()) {
+        n_ctx_train = common_get_gguf_n_ctx_train(params.model.path);
+    }
 }
 
 //
@@ -2125,6 +2130,10 @@ void server_models_routes::init_routes() {
                 // TODO: add other fields, may require reading GGUF metadata
             };
 
+            if (meta.n_ctx_train > 0) {
+                model_info["context_length"] = meta.n_ctx_train;
+            }
+
             // merge with loaded_info from the child process if available
             if (meta.is_running()) {
                 for (auto it = meta.loaded_info.begin(); it != meta.loaded_info.end(); ++it) {
diff --git a/tools/server/server-models.h b/tools/server/server-models.h
index 238955e7c4..f415b7c5fd 100644
--- a/tools/server/server-models.h
+++ b/tools/server/server-models.h
@@ -86,6 +86,7 @@ struct server_model_meta {
     int stop_timeout = 0; // seconds to wait before force-killing the model instance during shutdown
     bool hidden = false; // hidden from GET /models, but still accept if requested
     json architecture = server_model_architecture_json(false, false, false, {"text"});
+    uint32_t n_ctx_train = 0; // trained context, read from the GGUF metadata; 0 when unknown
 
     bool is_ready() const {
         return status == SERVER_MODEL_STATUS_LOADED;