diff --git a/misc/llama-cpp/Makefile b/misc/llama-cpp/Makefile
index 5effe760899b..61640736f856 100644
--- a/misc/llama-cpp/Makefile
+++ b/misc/llama-cpp/Makefile
@@ -1,50 +1,51 @@
 PORTNAME=	llama-cpp
 DISTVERSIONPREFIX=	b
-DISTVERSION=	4372
-PORTREVISION=	1
+DISTVERSION=	4381
 CATEGORIES=	misc # machine-learning
 
-PATCH_SITES=	https://github.com/${GH_ACCOUNT}/${GH_PROJECT}/commit/
-PATCHFILES=	a68c7eeec527b505e7c9226817e374e53e6cedb1.patch:-p1 # Fix run-time on FreeBSD in get_executable_path(): https://github.com/ggerganov/llama.cpp/pull/10948
-
 MAINTAINER=	yuri@FreeBSD.org
 COMMENT=	Facebook's LLaMA model in C/C++ # '
 WWW=		https://github.com/ggerganov/llama.cpp
 
 LICENSE=	MIT
 LICENSE_FILE=	${WRKSRC}/LICENSE
 
 BROKEN_armv7=	clang crashes, see https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=278810
 BROKEN_i386=	compilation fails, see https://github.com/ggerganov/llama.cpp/issues/9545
 
-USES=		cmake:testing compiler:c++11-lang python:run shebangfix
+USES=		cmake:testing compiler:c++11-lang python:run,test shebangfix
 USE_LDCONFIG=	yes
 
 USE_GITHUB=	yes
 GH_ACCOUNT=	ggerganov
 GH_PROJECT=	llama.cpp
 GH_TUPLE=	nomic-ai:kompute:4565194:kompute/kompute
 
 SHEBANG_GLOB=	*.py
 
 CMAKE_ON=	BUILD_SHARED_LIBS
 CMAKE_OFF=	LLAMA_BUILD_TESTS
 CMAKE_TESTING_ON=	LLAMA_BUILD_TESTS
 
 OPTIONS_DEFINE=		EXAMPLES VULKAN
 OPTIONS_DEFAULT=	VULKAN
 OPTIONS_SUB=		yes
 
 EXAMPLES_CMAKE_BOOL=	LLAMA_BUILD_EXAMPLES
 
 VULKAN_DESC=		Vulkan GPU offload support
 VULKAN_CMAKE_BOOL=	GGML_VULKAN
 VULKAN_BUILD_DEPENDS=	glslc:graphics/shaderc \
 			vulkan-headers>0:graphics/vulkan-headers
 VULKAN_LIB_DEPENDS=	libvulkan.so:graphics/vulkan-loader
 
-BINARY_ALIAS=	git=false
+BINARY_ALIAS=	git=false \
+		python3=${PYTHON_CMD}
+
+do-test-ci: # build of tests fails, see https://github.com/ggerganov/llama.cpp/issues/10955
+	@cd ${WRKSRC} && \
+		${SETENV} ${MAKE_ENV} bash ci/run.sh ./tmp/results ./tmp/mnt
 
-# 2 tests fail: https://github.com/ggerganov/llama.cpp/issues/8906
+# tests as of 4381: 97% tests passed, 1 tests failed out of 31
 
 .include <bsd.port.mk>
diff --git a/misc/llama-cpp/distinfo b/misc/llama-cpp/distinfo
index a9b2dbda376b..d9f5b907abca 100644
--- a/misc/llama-cpp/distinfo
+++ b/misc/llama-cpp/distinfo
@@ -1,7 +1,5 @@
-TIMESTAMP = 1734911761
-SHA256 (ggerganov-llama.cpp-b4372_GH0.tar.gz) = 9f3615c26da4a0f25123093d2d1a7822331777ba6bacea1de07e52a866901a49
-SIZE (ggerganov-llama.cpp-b4372_GH0.tar.gz) = 20594935
+TIMESTAMP = 1734927684
+SHA256 (ggerganov-llama.cpp-b4381_GH0.tar.gz) = 9fba75d7add548cd772c8edbccd00c11143e3f2b26155e12f33348a6ad8e695d
+SIZE (ggerganov-llama.cpp-b4381_GH0.tar.gz) = 20594961
 SHA256 (nomic-ai-kompute-4565194_GH0.tar.gz) = 95b52d2f0514c5201c7838348a9c3c9e60902ea3c6c9aa862193a212150b2bfc
 SIZE (nomic-ai-kompute-4565194_GH0.tar.gz) = 13540496
-SHA256 (a68c7eeec527b505e7c9226817e374e53e6cedb1.patch) = fd70161607513e10d95bb021c4413e47f5524c49897e56747235c4c4b5795202
-SIZE (a68c7eeec527b505e7c9226817e374e53e6cedb1.patch) = 1191