diff --git a/go.mod b/go.mod
index 90e902c..34935a4 100644
--- a/go.mod
+++ b/go.mod
@@ -5,7 +5,7 @@ go 1.23.7
 require (
 	github.com/containerd/containerd/v2 v2.0.4
 	github.com/containerd/platforms v1.0.0-rc.1
-	github.com/docker/model-distribution v0.0.0-20250418220155-bf3bed0d69b0
+	github.com/docker/model-distribution v0.0.0-20250418222450-8f6c05a5ffa4
 	github.com/jaypipes/ghw v0.16.0
 	github.com/opencontainers/go-digest v1.0.0
 	github.com/opencontainers/image-spec v1.1.1
diff --git a/go.sum b/go.sum
index 272537f..d48d6ef 100644
--- a/go.sum
+++ b/go.sum
@@ -37,8 +37,8 @@ github.com/docker/distribution v2.8.3+incompatible h1:AtKxIZ36LoNK51+Z6RpzLpddBi
 github.com/docker/distribution v2.8.3+incompatible/go.mod h1:J2gT2udsDAN96Uj4KfcMRqY0/ypR+oyYUYmja8H+y+w=
 github.com/docker/docker-credential-helpers v0.8.2 h1:bX3YxiGzFP5sOXWc3bTPEXdEaZSeVMrFgOr3T+zrFAo=
 github.com/docker/docker-credential-helpers v0.8.2/go.mod h1:P3ci7E3lwkZg6XiHdRKft1KckHiO9a2rNtyFbZ/ry9M=
-github.com/docker/model-distribution v0.0.0-20250418220155-bf3bed0d69b0 h1:p64u8SbT6qGcD5GcUkxkvjtm7LTHUpHUPtPE+7PBcTo=
-github.com/docker/model-distribution v0.0.0-20250418220155-bf3bed0d69b0/go.mod h1:fd9H/2KAY0+ByxbohWfCkxp0rxf0pqSlHEtFTjTXbLk=
+github.com/docker/model-distribution v0.0.0-20250418222450-8f6c05a5ffa4 h1:v7DyBUd08t4XQe4NC2Rb8ZSWOiWVXD0TbdgIOQZ/IQo=
+github.com/docker/model-distribution v0.0.0-20250418222450-8f6c05a5ffa4/go.mod h1:fd9H/2KAY0+ByxbohWfCkxp0rxf0pqSlHEtFTjTXbLk=
 github.com/felixge/httpsnoop v1.0.4 h1:NFTV2Zj1bL4mc9sqWACXbQFVBBg2W3GPvqp8/ESS2Wg=
 github.com/felixge/httpsnoop v1.0.4/go.mod h1:m8KPJKqk1gH5J9DgRY2ASl2lWCfGKXixSwevea8zH2U=
 github.com/go-logr/logr v1.2.2/go.mod h1:jdQByPbusPIv2/zmleS9BjJVeZ6kBagPoEUsqbVz/1A=
diff --git a/pkg/inference/models/manager.go b/pkg/inference/models/manager.go
index f795562..67f8592 100644
--- a/pkg/inference/models/manager.go
+++ b/pkg/inference/models/manager.go
@@ -8,6 +8,7 @@ import (
 	"html"
 	"net/http"
 	"path"
+	"strconv"
 	"strings"
 
 	"github.com/docker/model-distribution/distribution"
@@ -210,6 +211,8 @@ func (m *Manager) handleGetModel(w http.ResponseWriter, r *http.Request) {
 }
 
 // handleDeleteModel handles DELETE <inference-prefix>/models/{name} requests.
+// query params:
+// - force: if true, delete the model even if it has multiple tags
 func (m *Manager) handleDeleteModel(w http.ResponseWriter, r *http.Request) {
 	if m.distributionClient == nil {
 		http.Error(w, "model distribution service unavailable", http.StatusServiceUnavailable)
@@ -225,12 +228,24 @@ func (m *Manager) handleDeleteModel(w http.ResponseWriter, r *http.Request) {
 	// the runner process exits (though this won't work for Windows, where we
 	// might need some separate cleanup process).
 
-	err := m.distributionClient.DeleteModel(r.PathValue("name"))
-	if err != nil {
+	var force bool
+	if r.URL.Query().Has("force") {
+		if val, err := strconv.ParseBool(r.URL.Query().Get("force")); err != nil {
+			m.log.Warnln("Error while parsing force query parameter:", err)
+		} else {
+			force = val
+		}
+	}
+
+	if err := m.distributionClient.DeleteModel(r.PathValue("name"), force); err != nil {
 		if errors.Is(err, distribution.ErrModelNotFound) {
 			http.Error(w, err.Error(), http.StatusNotFound)
 			return
 		}
+		if errors.Is(err, distribution.ErrConflict) {
+			http.Error(w, err.Error(), http.StatusConflict)
+			return
+		}
 		m.log.Warnln("Error while deleting model:", err)
 		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return