llama.cpp b10417 fixes LFM2 tool calls with Claude Opus 5 help

llama.cpp, the wildly popular open-source C/C++ LLM inference engine with 124k stars and 21.7k forks on GitHub, released version b10417 on August 13. The headline change is a fix for "LFM2 tool call arg name prefix ambiguity" (issue #26960) — a subtle bug that could cause confusion when models using the LFM2 format generate tool calls with argument names that share prefixes, potentially leading to incorrect parameter parsing. Notably, the commit credits Claude Opus 5 for assistance, highlighting how AI is now co-authoring patches in major open-source projects.

The release ships as a tag b10417 with verified GPG signature, and includes prebuilt binaries across an enormous matrix of platforms and backends: macOS Apple Silicon (including a KleidiAI-enabled variant), macOS Intel, iOS XCFramework, multiple Linux builds (Ubuntu x64/arm64/s390x with CPU, Vulkan, ROCm 7.14, OpenVINO, SYCL FP32/FP16), Android arm64 CPU, Windows x64/arm64 with CPU, CUDA 12.4, CUDA 13.3, Vulkan, OpenVINO, SYCL, and ROCm 7.14, plus openEuler with Ascend NPU options. Some builds are marked DISABLED in this release, suggesting ongoing CI restructuring. The UI assets were also updated. For developers running local LLMs, this patch improves reliability in agent loops that rely on structured tool calls, enabling more robust multi-step reasoning with less manual intervention.