diff --git a/core/config/backend_capabilities.go b/core/config/backend_capabilities.go
index 79687c302..946527f4e 100644
--- a/core/config/backend_capabilities.go
+++ b/core/config/backend_capabilities.go
@@ -3,6 +3,8 @@ package config
 import (
 	"slices"
 	"strings"
+
+	"github.com/mudler/LocalAI/pkg/model"
 )
 
 // Usecase name constants — the canonical string values used in gallery entries,
@@ -718,6 +720,42 @@ func GetBackendCapability(backend string) *BackendCapability {
 	return nil
 }
 
+// llmAutoLoadUsecases are the usecases that mark a backend able to serve a
+// text/LLM GGUF model. A GGUF model that declares no explicit backend must only
+// be auto-tried against backends carrying one of these usecases - never against
+// audio/codec/image backends (e.g. opus) that happen to be installed alongside
+// it (see issue #9287).
+var llmAutoLoadUsecases = []string{
+	UsecaseChat,
+	UsecaseCompletion,
+	UsecaseEdit,
+	UsecaseEmbeddings,
+}
+
+// isLLMCapableForAutoLoad reports whether the named backend is known to serve
+// text/LLM models, for pkg/model's GGUF backend auto-detection (#9287). Backends
+// absent from the capability table are treated as not LLM-capable.
+func isLLMCapableForAutoLoad(name string) bool {
+	capability := GetBackendCapability(name)
+	if capability == nil {
+		return false
+	}
+	for _, u := range capability.PossibleUsecases {
+		if slices.Contains(llmAutoLoadUsecases, u) {
+			return true
+		}
+	}
+	return false
+}
+
+func init() {
+	// Wire the LLM-capability filter into pkg/model's GGUF backend
+	// auto-detection. pkg/model is a lower-level package and must not import
+	// core/config (that would form a core/config -> pkg/model -> core/config
+	// import cycle), so core/config registers the predicate here instead (#9287).
+	model.RegisterLLMCapableBackendFunc(isLLMCapableForAutoLoad)
+}
+
 // VoiceCloningForModel returns the reference-audio contract only when the
 // installed model variant can honor it. Several backends serve both Base
 // (voice cloning) and CustomVoice/VoiceDesign models, so backend name alone is
diff --git a/pkg/model/autoload.go b/pkg/model/autoload.go
index d6af3e095..d53186283 100644
--- a/pkg/model/autoload.go
+++ b/pkg/model/autoload.go
@@ -1,27 +1,28 @@
 package model
 
 import (
-	"slices"
 	"sort"
 	"strings"
-
-	"github.com/mudler/LocalAI/core/config"
 )
 
 // preferredGGUFBackend is tried first when auto-detecting the backend for a
 // GGUF model, since GGUF is overwhelmingly llama.cpp's native format.
 const preferredGGUFBackend = "llama-cpp"
 
-// llmCapableUsecases are the BackendCapabilities usecases that signal a backend
-// can serve a text/LLM GGUF model. A GGUF model that declares no explicit
-// backend must only be auto-tried against backends carrying one of these
-// usecases - never against audio/codec/image backends (e.g. opus) that happen
-// to be installed alongside it (see issue #9287).
-var llmCapableUsecases = []string{
-	config.UsecaseChat,
-	config.UsecaseCompletion,
-	config.UsecaseEdit,
-	config.UsecaseEmbeddings,
+// llmCapableBackend reports whether the named backend can serve a text/LLM GGUF
+// model. The backend capability table lives in core/config, which is a
+// higher-level package that already imports pkg/model; importing it back here
+// would form a core/config -> pkg/model -> core/config cycle. So core/config
+// registers the predicate via RegisterLLMCapableBackendFunc instead (see #9287).
+// When unset (e.g. a build that never imports core/config) GGUF capability
+// filtering is skipped and auto-detect falls back to the deterministic set.
+var llmCapableBackend func(name string) bool
+
+// RegisterLLMCapableBackendFunc wires the LLM-capability predicate used by
+// SelectAutoLoadBackends. It is called from core/config's init so pkg/model
+// need not import core/config (see #9287).
+func RegisterLLMCapableBackendFunc(fn func(name string) bool) {
+	llmCapableBackend = fn
 }
 
 // SelectAutoLoadBackends returns the ordered, deterministic list of backend
@@ -52,6 +53,12 @@ func SelectAutoLoadBackends(available []string, modelFile string) []string {
 		return sorted
 	}
 
+	// No capability predicate wired (core/config not linked in): skip filtering
+	// rather than risk dropping a valid candidate.
+	if llmCapableBackend == nil {
+		return sorted
+	}
+
 	filtered := make([]string, 0, len(sorted))
 	hasLlama := false
 	for _, b := range sorted {
@@ -84,16 +91,7 @@ func isGGUFModelFile(modelFile string) bool {
 // models. Backends absent from the capability map (unknown) are treated as
 // not LLM-capable here: for GGUF auto-detection we only want backends we can
 // positively confirm handle LLMs, and the zero-candidate fallback keeps unknown
-// setups working.
+// setups working. Callers must ensure llmCapableBackend is non-nil.
 func isLLMCapableBackend(name string) bool {
-	capability := config.GetBackendCapability(name)
-	if capability == nil {
-		return false
-	}
-	for _, u := range capability.PossibleUsecases {
-		if slices.Contains(llmCapableUsecases, u) {
-			return true
-		}
-	}
-	return false
+	return llmCapableBackend(name)
 }
diff --git a/pkg/model/autoload_test.go b/pkg/model/autoload_test.go
index f191969b4..9142639af 100644
--- a/pkg/model/autoload_test.go
+++ b/pkg/model/autoload_test.go
@@ -7,6 +7,15 @@ import (
 	. "github.com/onsi/gomega"
 )
 
+func init() {
+	// The real LLM-capability table lives in core/config, which pkg/model must
+	// not import (cycle). Register a fake predicate so the selection algorithm
+	// is exercised here independent of the capability table (#9287).
+	model.RegisterLLMCapableBackendFunc(func(name string) bool {
+		return name == "llama-cpp" || name == "vllm"
+	})
+}
+
 var _ = Describe("SelectAutoLoadBackends (#9287)", func() {
 	Describe("GGUF model auto-detection", func() {
 		It("excludes incompatible audio/codec backends (e.g. opus) for a .gguf model", func() {