diff --git a/misc/llama-cpp/Makefile b/misc/llama-cpp/Makefile
index c314d14b277e..26624d563682 100644
--- a/misc/llama-cpp/Makefile
+++ b/misc/llama-cpp/Makefile
@@ -1,81 +1,81 @@
 PORTNAME=	llama-cpp
 DISTVERSIONPREFIX=	b
-DISTVERSION=	5862
+DISTVERSION=	5891
 CATEGORIES=	misc # machine-learning
 
 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
 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=	GGML_NATIVE \
 		FREEBSD_ALLOW_ADVANCED_CPU_FEATURES \
 		LLAMA_BUILD_TESTS
 CMAKE_TESTING_ON=	LLAMA_BUILD_TESTS
 
 # user for llama-server, only used when EXAMPLES=ON
 USER=		nobody
 SUB_LIST=	USER=${USER}
 
 OPTIONS_DEFINE=		CURL EXAMPLES VULKAN
 OPTIONS_DEFAULT=	CURL VULKAN
 OPTIONS_SUB=		yes
 
 CURL_DESCR=		Use libcurl to download model from an URL
 CURL_CMAKE_BOOL=	LLAMA_CURL
 CURL_USES=		localbase
 CURL_LIB_DEPENDS=	libcurl.so:ftp/curl
 
 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 \
 		python=${PYTHON_CMD} # for tests
 
 post-patch: # set version in the code
 	@${REINPLACE_CMD} \
 		-e "s|set(BUILD_NUMBER 0)|set(BUILD_NUMBER ${DISTVERSION})|" \
 		${WRKSRC}/cmake/build-info.cmake
 
 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
 
 .include <bsd.port.options.mk>
 
 .if ${PORT_OPTIONS:MEXAMPLES}
 USE_RC_SUBR=	llama-server
 .endif
 
 # tests as of 4458: 97% tests passed, 1 tests failed out of 31, see https://github.com/ggerganov/llama.cpp/issues/11036
 
 # tests as of 4649:
 # 88% tests passed, 4 tests failed out of 32
 # The following tests FAILED:
 #         18 - test-chat (Subprocess aborted)                    main   # see https://github.com/ggerganov/llama.cpp/issues/11705
 #         24 - test-gguf (SEGFAULT)                              main
 #         25 - test-backend-ops (SEGFAULT)                       main
 #         32 - test-eval-callback (SEGFAULT)                     curl eval-callback
 
 .include <bsd.port.mk>
diff --git a/misc/llama-cpp/distinfo b/misc/llama-cpp/distinfo
index 49c7c2cacb3d..61d48f40a276 100644
--- a/misc/llama-cpp/distinfo
+++ b/misc/llama-cpp/distinfo
@@ -1,5 +1,5 @@
-TIMESTAMP = 1752140413
-SHA256 (ggerganov-llama.cpp-b5862_GH0.tar.gz) = 987e8e7a58a38e30c0966c59f65318c7e18506685f580bf70dcda2dc892aa47f
-SIZE (ggerganov-llama.cpp-b5862_GH0.tar.gz) = 25043505
+TIMESTAMP = 1752483190
+SHA256 (ggerganov-llama.cpp-b5891_GH0.tar.gz) = c3475547d5036ba2afa9467e6adbc6c617f5eba7d6256a96fa8c8f2206429842
+SIZE (ggerganov-llama.cpp-b5891_GH0.tar.gz) = 25176497
 SHA256 (nomic-ai-kompute-4565194_GH0.tar.gz) = 95b52d2f0514c5201c7838348a9c3c9e60902ea3c6c9aa862193a212150b2bfc
 SIZE (nomic-ai-kompute-4565194_GH0.tar.gz) = 13540496