{"id":16153,"date":"2026-11-17T11:04:33","date_gmt":"2026-11-17T02:04:33","guid":{"rendered":"https:\/\/localmodelwatch.tsuchitsuchi.com\/?p=16153"},"modified":"2026-11-17T11:13:59","modified_gmt":"2026-11-17T02:13:59","slug":"llamacpp-v100-released","status":"publish","type":"post","link":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/llamacpp-v100-released\/","title":{"rendered":"llama.cpp v1.0.0 Released: Major Updates and New Features"},"content":{"rendered":"<p><!-- lmw:facts --><\/p>\n<h2>At a Glance<\/h2>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Item<\/th>\n<th>Value<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Repository<\/td>\n<td><a href=\"https:\/\/github.com\/ggml-org\/llama.cpp\">ggml-org\/llama.cpp<\/a><\/td>\n<\/tr>\n<tr>\n<td>Version<\/td>\n<td><a href=\"https:\/\/github.com\/ggml-org\/llama.cpp\/releases\/tag\/v1.0.0\">v1.0.0<\/a><\/td>\n<\/tr>\n<tr>\n<td>Published<\/td>\n<td>2026-11-13<\/td>\n<\/tr>\n<tr>\n<td>License<\/td>\n<td>MIT<\/td>\n<\/tr>\n<tr>\n<td>Source type<\/td>\n<td>Primary source (the publisher itself)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><em>Values determined by this site&#8217;s code when the information was collected. Dates are JST.<\/em><\/p>\n<p><!-- \/lmw:facts --><\/p>\n<h2>Overview<\/h2>\n<p>llama.cpp is an LLM inference engine written in C\/C++ (MIT license). Version 1.0.0 has been released. The official note simply states &#8220;More info at 3PM CET&#8221;, and detailed explanations are expected to be added later. The following is based on the commit list since v0.9.0, and details remain unconfirmed.<\/p>\n<p>The change with the most impact for users is the security hardening of <code>llama-server<\/code>. It includes fixes for <code>--cors-origins<\/code> processing and DNS rebinding mitigation (#32098), as well as SSRF fixes for CORS proxies and media URLs (#32115). Users exposing servers to LANs or other networks should consider updating promptly.<\/p>\n<h2>Key Changes<\/h2>\n<p>Commit titles are the only basis, as PR descriptions are not available in the source material. Therefore, specific details of behaviors cannot be definitively confirmed.<\/p>\n<h3>New Features<\/h3>\n<p><strong>FP8 KV-cache and CUDA OCP FP8 Support<\/strong><\/p>\n<p><code>llama : Add FP8 KV-cache support<\/code>(#28392) and <code>CUDA: Add OCP FP8 support<\/code>(#28413) have been added. This targets CUDA users looking to save VRAM. Usage instructions and supported GPUs are not in the documentation, pending official follow-up.<\/p>\n<p><strong>Windows Unbuffered Model Load<\/strong><\/p>\n<p><code>Windows unbuffered model load<\/code>(#26014) has been added. This is relevant for users loading large models on Windows. Instructions on how to enable it are not documented.<\/p>\n<p><strong>External PLE File Loading<\/strong><\/p>\n<p><code>feat: external PLE file loading<\/code>(#31185) has been added. Details are not provided in the documentation.<\/p>\n<p><strong>Server Audio and MCP Related<\/strong><\/p>\n<ul>\n<li>OpenAI-compatible <code>\/v1\/audio\/speech<\/code> endpoint has been added(#31763)<\/li>\n<li>MCP tool calls are now cancelled on client disconnection or timeout(#32025)<\/li>\n<li>When <code>--tools-runtime<\/code> is specified, <code>x-tool-runtime<\/code> is ignored(#32119)<\/li>\n<\/ul>\n<p><strong>MoE Cache Improvements<\/strong><\/p>\n<p>An LRFU\/MQ-style hybrid caching method(#31233) has been introduced. Processing to execute experts not present in the cache on the CPU(#30966) has also been added. Users running MoE models that do not fit entirely in VRAM are advised to verify the behavior.<\/p>\n<h3>Performance Improvements<\/h3>\n<ul>\n<li>CUDA: Few-row MMQ for batch sizes 2\u201324 (#31533)<\/li>\n<li>CUDA: Optimized nvfp4 checkpoint prefill with scale fusion(#26048)<\/li>\n<li>Vulkan: FlashAttention kernel for prefill targeting Intel(#29357)<\/li>\n<li>ARM NEON: Utilization of 4&#215;4 simd_gemm tiles(#32128)<\/li>\n<li>Hexagon: Support for Q5_0 and Q5_1(#32094)<\/li>\n<\/ul>\n<p>Specific numerical performance gains are not included in the documentation.<\/p>\n<h3>Bug Fixes and Security<\/h3>\n<ul>\n<li>A change has been introduced where <code>llama-server<\/code> treats tool usage on non-loopback addresses as requiring explicit opt-in(#32113). In externally exposed environments, the behavior of tool functions may change (details unconfirmed)<\/li>\n<li>Restricted Hugging Face tokens to be sent only to the huggingface.co host(#31730)<\/li>\n<li>Fixed an issue where llama-cli would crash on invalid UTF-8 in prompts or files(#31563)<\/li>\n<li>Fixed out-of-bounds writes to Metal FlashAttention output buffers(#31662)<\/li>\n<li>Fixed an issue in <code>llama-quantize<\/code> where imatrix GGUF file paths leaked into metadata(#32000)<\/li>\n<li>In mtmd, strict validation of mandatory tensors was added for projectors such as Qwen-VL and HunyuanVL(#32129, #32130, #32131). If your existing mmproj fails to load, try re-converting or re-acquiring it (unconfirmed)<\/li>\n<\/ul>\n<h2>Supported Models and Hardware<\/h2>\n<p>This version brings support for multiple new model architectures, various accelerators, and quantization formats.<\/p>\n<h3>Newly Supported Models and Extensions<\/h3>\n<p>According to the documentation&#8217;s commit list, the following model architectures and feature additions have been reported:<\/p>\n<ul>\n<li><strong>Kimi-K3.1<\/strong>: Support added for the text model <code>Kimi-K3.1-Pro<\/code>(#31779) and <code>kimi K3.1 vision<\/code>(#31980).<\/li>\n<li><strong>Gemma 4<\/strong>: In addition to schema rule organization(#31915), support for input audio streams(#31965) has been added.<\/li>\n<li><strong>TTS (Text-to-Speech) Models<\/strong>: Support and feature extensions implemented for <code>drama-3<\/code> TTS(#31613) and <code>Pocket TTS<\/code>(#31655).<\/li>\n<li><strong>Other Models<\/strong>: Added support for models such as <code>Ripple (ricursiveai)<\/code>(#31692), <code>xing4_0<\/code>(#31589), and <code>MageVL-Cowork<\/code>(#31819). Byte fallback token processing fixes for <code>Sarvam<\/code>(#32006) are also included.<\/li>\n<\/ul>\n<p>In multimodal (mtmd) processing, mandatory F32 positional encoding has been introduced for <code>HunyuanVL<\/code>(#32131), along with improved projection width loading for <code>MiniCPM-V<\/code>(#32132) and the requirement for 2&#215;2 spatial merging in the <code>Qwen-VL<\/code> projector(#32130), streamlining inference processing for related models.<\/p>\n<h3>Quantization and Accelerator Support<\/h3>\n<p>Optimizations and expansions have also been carried out across many backends for hardware accelerators and quantization formats.<\/p>\n<ul>\n<li><strong>Hexagon<\/strong>: Added support for <code>Q5_0<\/code> and <code>Q5_1<\/code> quantization formats(#32094).<\/li>\n<li><strong>CUDA \/ GPU<\/strong>: Implemented OCP FP8 format support(#28413), scale fusion optimization for nvfp4 checkpoints(#26048), and MFMA indexer for CDNA2 (gfx90a)(#29050). Weight scale attachment for Granite NVFP4(#31449) is also supported.<\/li>\n<li><strong>Vulkan<\/strong>: Added FlashAttention kernel for prefill targeting Intel(#29357) and Large-Tile Split-K optimizations for AMD RDNA3\/4(#30280).<\/li>\n<li><strong>OpenCL \/ Mobile<\/strong>: Added decomposed FlashAttention for Qualcomm Adreno(#32046), and multi-column GEMV for F32 and Q5_K on Adreno X2 enabled by default(#31694).<\/li>\n<li><strong>SpacemiT \/ RISC-V<\/strong>: Added <code>GATED_DELTA_NET<\/code> routing to SpacemiT RVV kernels(#31897), SpacemiT Q4_1 32&#215;256 HP kernel dispatch(#31929), and enabling of SpacemiT IME2 MXFP4 Dense Repack(#31776).<\/li>\n<li><strong>SYCL \/ WebGPU<\/strong>: Added device-side MoE scheduler in the SYCL backend(#30156), and implemented <code>PAD_REFLECT_1D<\/code>(#31877), <code>COUNT_EQUAL<\/code>(#31578), <code>MEAN<\/code>(#31591), and Group Norm kernels(#31490) in WebGPU.<\/li>\n<\/ul>\n<h2>How to Get It<\/h2>\n<p>The documentation&#8217;s build guide does not list specific installation commands. For environment-specific instructions on obtaining the latest binaries or update procedures, please refer directly to the <a href=\"https:\/\/github.com\/ggml-org\/llama.cpp\/releases\/tag\/v1.0.0\">llama.cpp releases page<\/a>.<\/p>\n<p>If updating from source code, fetch the latest code using standard git commands and rebuild:<\/p>\n<pre><code class=\"language-bash\">git pull\ncmake -B build\ncmake --build build --config Release\n<\/code><\/pre>\n<p><!-- lmw:version-speed --><\/p>\n<h2>Speed on Our CPU<\/h2>\n<p><em>Measured by Local Model Watch on our CPU-only server (Neoverse-N1, 3 threads) with llama-bench (512-token prompt, 128-token generation), always on the same model: <code>Qwen3-0.6B-Q4_K_M.gguf<\/code> from <a href=\"https:\/\/huggingface.co\/unsloth\/Qwen3-0.6B-GGUF\/tree\/50968a4468ef4233ed78cd7c3de230dd1d61a56b\">unsloth\/Qwen3-0.6B-GGUF<\/a> (revision <code>50968a4468<\/code>). Versions like v0.5.0 ship no binaries, so we used the first official ubuntu-arm64 build that contains the version; \u201c+N\u201d is how many commits that build is ahead of it. These are CPU figures, not a guide to GPU speed, and small differences can be run-to-run noise (\u00b1 is the standard deviation).<\/em><\/p>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Version<\/th>\n<th>Build measured<\/th>\n<th>Prompt<\/th>\n<th>Generation<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>v1.0.0<\/td>\n<td>b12359<\/td>\n<td>132.7 \u00b1 0.6 tok\/s<\/td>\n<td>42.5 \u00b1 0.7 tok\/s<\/td>\n<\/tr>\n<tr>\n<td>v0.9.0<\/td>\n<td>b12110<\/td>\n<td>128.1 \u00b1 0.9 tok\/s<\/td>\n<td>42.9 \u00b1 0.6 tok\/s<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><!-- \/lmw:version-speed --><\/p>\n<p><!-- lmw:related --><\/p>\n<h2>Related Articles<\/h2>\n<ul>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/llama-cpp-v0-9-0-released\/\">llama.cpp v0.9.0 Released: FP8 Quantization and More<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/10\/06\/llamacpp-v0-6-0-released\/\">llama.cpp v0.6.0 Released with llama_batch_ext and Metal Boosts<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/09\/24\/llama-cpp-v0-5-0-released\/\">llama.cpp v0.5.0 Released with Backend and Server Upgrades<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/llama-cpp-v0-8-0-multimodal-lfm3-support\/\">llama.cpp v0.8.0: Multimodal Enhancements and LFM3 Support<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:related --><\/p>\n<p><!-- lmw:next-steps --><\/p>\n<h2>What to Read Next<\/h2>\n<ul>\n<li><strong>Follow this tool<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-llama-cpp-en\/\">llama.cpp overview and release history (9 releases tracked)<\/a><\/li>\n<li><strong>Other inference engines and runtimes<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-ollama-en\/\">Ollama<\/a> \/ <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-vllm-en\/\">vLLM<\/a> \/ <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-sglang-en\/\">SGLang<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:next-steps --><\/p>\n<h2>Sources<\/h2>\n<ul>\n<li><a href=\"https:\/\/github.com\/ggml-org\/llama.cpp\/releases\/tag\/v1.0.0\">ggml-org\/llama.cpp v1.0.0<\/a><\/li>\n<li>[Releases and versioning of ggml-org projects]<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>llama.cpp v1.0.0 has been released. Read about security enhancements, FP8 KV-cache, new model support, hardware accelerations, and key changes.<\/p>\n","protected":false},"author":1,"featured_media":16152,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[105],"tags":[1187,1133,163,1073,165,1547],"class_list":["post-16153","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-engines-and-tools","tag-cuda-en","tag-fp8-en","tag-gguf-en","tag-llama-cpp-en","tag-moe-en","tag-verified"],"lang":"en","translations":{"en":16153,"ja":16151},"_links":{"self":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/16153","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/comments?post=16153"}],"version-history":[{"count":2,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/16153\/revisions"}],"predecessor-version":[{"id":18118,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/16153\/revisions\/18118"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media\/16152"}],"wp:attachment":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media?parent=16153"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/categories?post=16153"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/tags?post=16153"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}