llama.cpp v0.6.0: llama_batch_ext API, GLM-5.3-Flash and /v1/systemone
llama.cpp v0.6.0 adds the llama_batch_ext batch API (llama_process) for mixed token/embedding inputs, supports GLM-5.3-Flash 320B and Clef models, adds /v1/systemone server API, Metal kernels, and ggml v0.26.0.
What changed
Added llama_batch_ext API with llama_process; added GLM-5.3-Flash 320B and Clef models; new /v1/systemone server API; Metal F16 KV flash attention kernel; updated ggml to v0.26.0.
Why it matters
Developers can use mixed token/embedding batches and MTP/deepstack state embeddings, run GLM-5.3-Flash and Clef text+vision models, and deploy decision models via the new /v1/systemone API.
Who should care
DevelopersResearchersEnterprises
Worth trying
Yes — Use llama_batch_ext for mixed token/embedding inputs and new vision models.