Agentic WeeklyGet the weekly
ggml-org (open-source project)b11889Tested (smoke test)

llama.cpp: what’s new in b11889 and our verdict

LLM inference in C/C++: the open-source engine many local-model tools build on; MIT-licensed with near-daily build tags.

Official site

Release Ruler for llama.cpp: marked at its Tested release date on the timeline of every AI model we track.

At a glance

Evidence
Tested (smoke test)
Released
Oct 25, 2026
Last verified
Oct 25, 2026
Pricing
Official page, dated
Best for
local LLM tools and model platforms

Evidence and ranking

Evidence behind this verdict

  1. Sources read1 cited source, latest fetched Oct 25, 2026. (done)
  2. Facts re-checkedLast verified Oct 25, 2026. (done)
  3. Run by usSmoke test on Oct 25, 2026 (scope below). (done)

Passed Test log result: Passed. Read the log

Where we rank it (editorial)

Listed here but not yet ranked. Ranks are editorial picks, not benchmark scores.

Pricing: Open-source project; no pricing. (checked 2026-10-25)

Our verdict

LLM inference in C/C++: the open-source engine many local-model tools build on; MIT-licensed with near-daily build tags.

Good for

  • Developers embedding local inference
  • People who want maximum control over local models

Skip if

  • You want a polished app with a GUI

Tested. Tested 2026-10-25 (America/Chicago). llama.cpp b11889 generated 118 tokens/s (0.5B), 40 (1.5B) and 19.5 (3B) on 8 CPU threads with llama-bench, and its server passed the same prompts and tool call as Ollama. Test environment: Linux x86_64 virtual machine (Debian 13), 8 vCPU, 15 GiB RAM, no GPU; no vendor account, login or API key. Agent tools were pointed at a local open-weight model (Qwen2.5-3B-Instruct (Q4_K_M) through Ollama 0.40.11), so we tested the tool, not any vendor's hosted models. Not tested: Building from source, GPU backends, longer prompts, quantisations other than Q4_K_M, speculative decoding; Single bench run of 64 tokens; it is a smoke test and not a benchmark suite. These are smoke tests, not a benchmark. Full log: /tests/llama-cpp.

Full test log with exact commands, versions and raw output: llama.cpp b11889 (commit d854d98ba): our test log.

Evidence
Tested (smoke test, 2026-10-25, local 3B model (Qwen2.5-3B)). A smoke test is not a benchmark.
What we ran
llama.cpp b11889 generated 118 tokens/s (0.5B), 40 (1.5B) and 19.5 (3B) on 8 CPU threads with llama-bench, and its server passed the same prompts and tool call as Ollama.
Model used
Qwen2.5-3B-Instruct (Q4_K_M) through Ollama 0.40.11
Environment
Linux x86_64 virtual machine (Debian 13), 8 vCPU, 15 GiB RAM, no GPU; no vendor account, login or API key
Not tested
Building from source, GPU backends, longer prompts, quantisations other than Q4_K_M, speculative decoding; Single bench run of 64 tokens; it is a smoke test and not a benchmark suite.
Date
(America/Chicago)

Best for

Key facts

The llama.cpp repository describes LLM inference in C/C++ and is MIT-licensed. Its releases are build tags, with b11889 published Oct 25, 2026.

  • License: MIT (per GitHub)
  • Platforms: Cross-platform source and binaries
  • Pricing model: Free and open source
  • Free tier: Free
  • Paid plans: None
  • GitHub: ggml-org/llama.cpp: 132,360 stars on 2026-10-25, last push 2026-10-26
  • Latest release we saw: b11889, 2026-10-25
  • How to start: Clone the repository or download a release build and follow the README's quick start.
  • b11889 (2026-10-25): the newest release listed on the official page or repository when we read it on 2026-10-25.

How to set it up: Get started with llama.cpp

These steps come from the vendor's own pages as read on 2026-10-25.cpp. We ran llama.cpp on 2026-10-25; the test log lists what was and was not covered.

  1. Get access

    Clone the repository or download a release build and follow the README's quick start.

  2. Check the limits and costs first

    Free. Open-source project; no pricing.

Pros and cons

Pros

  • MIT license
  • Very active

Cons

  • Command-line and build-oriented
  • Our test was one small task on a local 3B model; it does not show behaviour with hosted models or larger jobs

Alternatives and comparisons

  • Ollama: The common way to run open models locally on macOS, Windows and Linux, with a cloud option and an MIT-licensed open-source core.
  • LM Studio: Desktop app for running open models locally; the homepage now leads with Bionic, an agent for work and code.
  • Jan: Open-source alternative to ChatGPT that runs open models locally or connects to cloud models like GPT and Claude.
  • Cohere: Enterprise AI company: Command models, Rerank and the North platform.

FAQ

What is llama.cpp?

The llama.cpp repository describes LLM inference in C/C++ and is MIT-licensed. Its releases are build tags, with b11889 published Oct 25, 2026.

Is llama.cpp free?

Free. Open-source project; no pricing. Checked 2026-10-25; prices change, so confirm on the vendor's page.

What license is llama.cpp under?

Its license: MIT (per GitHub). Read from the project's own pages on 2026-10-25.

What we have not verified

  • Performance on your hardware

Sources

Last verified · Agentic Weekly

  • Cohere: Enterprise AI company: Command models, Rerank and the North platform.
  • DeepSeek: Model API that works with the OpenAI and Anthropic SDK formats and can back agent tools such as Claude Code, GitHub Copilot and OpenCode.
  • Hugging Face: The hub where the ML community shares models, datasets and apps, with PRO, Team and Enterprise tiers.

How to cite this page

Agentic Weekly, "llama.cpp b11889: what's new, verdict & setup", last updated 2026-10-25, https://theagenticweekly.com/tools/llama-cpp

Keep the Reported or Tested label when you quote a verdict. Plain-text version: https://theagenticweekly.com/tools/llama-cpp.md. More for AI assistants: /ai, /tools.json, /llms.txt.