ggml-org/llama.cpp build b11768 was released on October 21, 2026. It is a build in which the server of this project, which implements LLM inference in C and C++, was fixed to include the output item number (output_index) that had been missing from Responses API streaming responses. The source of the change is the b11768 release on GitHub.
The Responses API is a specification that returns model responses in a defined format, and Server-Sent Events (SSE) is a method that does not send the response all at once but splits it into pieces and streams them in sequence. output_index is the number that tells which output item a given piece belongs to. This change made the server include this previously missing value in each piece.
The repository's public star count is 131855. The same build reused the Anthropic streaming code, and two entries, generated_tool_call_ids and oai_resp_tool_calls_started, which were identified as no longer needed, were removed.
However, the change is limited to the Responses API streaming path. Because it is a release at the build-number level, the given information says nothing about its impact on overall performance or on other API paths. No test results or benchmarks are presented either, so it is too early to judge whether streaming behavior in other response formats improved as well. b11768 is best understood as a narrow fix that adds the output item number, and there is no basis to read it as a build that brought changes in performance or other features.