b9980

ggml-org/llama.cppb9980Jul 13, 2026by github-actions[bot]same day as b9979

What changed since b9979

low risk

Server now reports models with mmproj explicitly disabled as text-only on /v1/models instead of falsely advertising multimodal input.

Improvements

  • The /v1/models endpoint now derives supported input modalities from the actual mmproj state rather than assuming multimodal capability.
  • Model presets and command-line flags that disable mmproj are now honored consistently in API capability reporting.

Bug fixes

  • Fixed /v1/models incorrectly listing image and audio as supported input modalities for models where mmproj was turned off via --no-mmproj-auto or a preset.

Upgrade notes

This is a server-side reporting fix only; no configuration or migration steps are required. Clients that relied on the previous (incorrect) multimodal advertisement for models launched with mmproj disabled will now see text-only modalities, so verify any capability-detection logic against the corrected output.

Analyzed Oct 24, 2026

Upgrading?

Release notes summary

Fixes the `/v1/models` endpoint to correctly report input modalities, ensuring that models with multimodal projection explicitly disabled via `--no-mmproj-auto` are not falsely advertised as multimodal.

Key Highlights

  • API endpoint accuracy
  • Respects `--no-mmproj-auto` flag
  • Correct multimodal reporting

New Features

  • Server API accuracy

Release Notes (excerpt)

<details open> server: Don't consider models with --no-mmproj-auto as multimodal (#25590) If mmproj is explicitly disabled via the model preset or command-line parameters then the model won't be able to handle image/audio inputs and this shouldn't be declared as supported input modality on the /v1/models endpoint. </details> **macOS/iOS:** - macOS Apple Silicon (arm64) - macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED - macOS Intel (x64) - iOS XCFramework **Linux:**…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories