lemonade-sdk/lemonade candidate-v2026.42.1 on GitHub

Headline

Streaming requests no longer hang when a client disconnects mid-request.
Windows support added for TheNoise (ROCm) via a bundled portable CPython interpreter.
llama.cpp backend updated with six new model architectures (clef, glm5-next, hrm_text, hy_v4, maple, spark2_5) and HRX runtime updated to hrx-b99 with 26 new text-chat models.
extra_models_dir now correctly import models from Hugging Face style cache directories.

The FastFlowLM NPU model hy-mt2:1.8b was renamed to hy-mt2-flash:1.8b; update any configs or scripts that reference the old tag.

Operating System	Downloads
Windows	lemonade.msi
Ubuntu 24.04+	Launchpad PPA
Debian 13 (x86_64)	lemonade-server_2026.42.1-debian13_amd64.deb
Debian 13 (ARM64)	lemonade-server_2026.42.1-debian13_arm64.deb
Fedora 43 (x86_64)	lemonade-server-2026.42.1-fc43.x86_64.rpm
Fedora 43 (ARM64)	lemonade-server-2026.42.1-fc43.aarch64.rpm
Fedora 44 (x86_64)	lemonade-server-2026.42.1-fc44.x86_64.rpm
Fedora 44 (ARM64)	lemonade-server-2026.42.1-fc44.aarch64.rpm
macOS	Lemonade-2026.42.1-Darwin.pkg

Other platforms? See our Installation Options for Docker, Snap, Arch, Debian, and more.

Portable binaries for bundling into your own installer. Run lemond ./ as a subprocess.

Platform	Download
Ubuntu x64	lemonade-embeddable-2026.42.1-ubuntu-x64.tar.gz
Ubuntu arm64	lemonade-embeddable-2026.42.1-ubuntu-arm64.tar.gz
Windows x64	lemonade-embeddable-2026.42.1-windows-x64.zip
macOS arm64	lemonade-embeddable-2026.42.1-macos-arm64.tar.gz

Thanks @AaronStGeorge, @bitgamma, @github-actions, @jeremyfowers, @julianxhokaxhiu, @ramkrishna2910 for your awesome contributions to this release!

Click to expand changelog

Auto-install specified backends with lemonade bench by @bitgamma in #3690
test: remove flaky test_000b_retired_endpoints_removed by @jeremyfowers in #3720
Add Windows support for TheNoise by @bitgamma in #3666
Don't fail HTTPS requests on Windows when cert revocation status is unknown by @ramkrishna2910 in #3716
fix(models): group shard files at the extra_models_dir root by @jeremyfowers in #3656
fix(server): watch extra_models_dir's nested folders on Linux and macOS by @jeremyfowers in #3726
docs(models): specify extra_models_dir behavior in the custom model guide by @jeremyfowers in #3633
test(models): describe extra_models_dir layouts declaratively by @jeremyfowers in #3634
feat(hrx): bump hrx to b99 + add 26 models by @AaronStGeorge in #3723
chore(backends): bump FastFlowLM to v1.0.7 by @julianxhokaxhiu in #3744
fix(http): make download stall timeouts follow global_timeout by @jeremyfowers in #3730
fix(models): give a shared folder name to one model in both forms by @jeremyfowers in #3657
Change evaluation date for AI content policy by @jeremyfowers in #3760
bump TheNoise version by @bitgamma in #3750
feat(models): support Hugging Face and ModelScope caches in extra_models_dir by @jeremyfowers in #3635
fix(tray): make --spawn-server a no-op on macOS by @ramkrishna2910 in #3756
docs: add the container backends spec by @jeremyfowers in #3697
Update llama.cpp to b11393 by @github-actions[bot] in #3754

Full Changelog: v2026.41.1...candidate-v2026.42.1

The Windows installers in this release candidate are not signed; Windows SmartScreen will warn before installing. Stable releases are signed.