Repo Compare › LLM inference

vLLM vs llama.cpp vs Text Generation Inference

A side-by-side look at how actively vLLM, llama.cpp and Text Generation Inference are developed on GitHub — popularity, contributor base, commit activity and how quickly issues and pull requests move.

What the numbers say

• llama.cpp is the most-starred of the 3, with 130k stars — about 11.9× Text Generation Inference's 11k.
• Over the 52 weeks to the snapshot, vLLM recorded the most commits (12k), against 8 for Text Generation Inference.
• vLLM has closed or merged the most pull requests in its lifetime: 34k.
• Relative to its popularity, llama.cpp carries the lightest open-issue load: 6.7 open issues per 1,000 stars, versus 27.4 for vLLM.
• Text Generation Inference had gone 193 days without a push at the time of the snapshot.
• Text Generation Inference is archived on GitHub — read-only, no longer accepting changes.

Head-to-head

Snapshot taken 30 September 2026. Refresh with live data

The wide bar is that repository’s share of the largest value in its own row — half the length means half the number. The thin gauge beneath it is the same row rescaled from its smallest value to its largest, which makes a close gap such as 61k against 59k visible; on that gauge the row’s lowest value always sits at zero however big it is, so read it as a spread and never as a quantity. Bars compare repositories with each other and never one metric against another. Colour identifies the repository, matching the dot beside its name. On the two issue rows a longer bar means more open work, which is usually the opposite of better. The highlighted cell is the highest value in a row. “Active / Slowing / Dormant” is a simple heuristic from how recently code was pushed and open issues per star — a starting point, not a verdict. Stars measure popularity, not quality.

Open these in the live comparer →

Related comparisons

• Ollama vs llama.cpp vs LocalAI
• LangChain vs LlamaIndex vs Haystack
• AutoGen vs CrewAI vs LangGraph