Model components / Inference code

llamafile

llamafile v0.10.3 (released June 2, 2026, GitHub); Apache-2.0 (MIT for llama.cpp/whisper.cpp changes). Single-file cross-platform LLM executable (llama.cpp + Cosmopolitan Libc). Mozilla Builders initiative, actively maintained June 2026.

Openness

5 high confidence

5.0

license: Apache-2.0(OSI)+MIT(OSI, for llama.cpp/whisper.cpp deltas)
source: public(Mozilla-Ocho/llamafile)
core-gated: ungated

Fully OSI-licensed, source public, single-file llama.cpp+Cosmopolitan distribution; no proprietary tier.

https://github.com/Mozilla-Ocho/llamafile recorded 2026-06-04
Apache-2.0 (+ MIT deltas) license, v0.10.3 (June 2 2026), single-file executable, 24.7k stars, actively maintained

Adoption

3 low confidence

3.0

No PyPI/download or pull-count signal available (distributed as single-file binaries, not a package); 24.7k GitHub stars + Mozilla backing is the only honest signal. Capped at 3 per stars_fallback rule. Popular local-inference convenience tool but no verified usage_volume figure.

https://github.com/Mozilla-Ocho/llamafile recorded 2026-06-04
24.7k stars, 40 releases, Mozilla Builders initiative

Capability

3 medium confidence

3.0

Optimizes for portability/zero-install local inference rather than datacenter throughput; built on llama.cpp so no continuous-batching/TP server scaling like vLLM. Mid-tier on a capability axis whose frontier (C5) is the high-throughput serving engines.

https://github.com/Mozilla-Ocho/llamafile recorded 2026-06-04
single-file llama.cpp + Cosmopolitan Libc, cross-OS, whisperfile STT

Unchanged since 2026-07-30 (last edited, not re-checked)