llama.cpp adds support for Thinking Machines’ Inkling 2 text and vision
5.0/10Significance
On November 15, 2026 the llama.cpp project published pre-release b12406 with support for Inkling 2, covering text and vision. The release notes say real checkpoint conversion remains untested.
What changed
The release adds Inkling 2 text GGUF conversion and a still-image encoder. The notes report CPU logits within 3.1e-6 absolute error on packed synthetic weights.
Why it matters
It is a step toward running the open Inkling 2 Small weights in common local inference tools.
What remains uncertain
The notes say real checkpoint conversion is untested, and the build is marked pre-release.
What to watch
Watch for a follow-up that tests conversion on the real released checkpoint.
Source
- github.comRelease b12406 · ggml-org/llama.cpp
Why this ranks here
A useful open-source tooling step, but early and unverified on real weights.
- Impact4
- Reach5
- Novelty5
- Commitment5
- Coverage quality5
- Future potential6