lemonade-sdk/lemonade candidate-v2026.40.0 on GitHub

Headline

Streaming models now run on AMD integrated GPUs by sizing against the APU GTT pool instead of the fixed VRAM carve-out, enabling model streaming on hardware previously unsupported.
Junie by JetBrains is a new launch agent available via lemonade launch junie, with profile generation and integration documentation.
Cloud provider context window and completion token limits are now parsed from model metadata and surfaced on GET /v1/models.
AMD GPU entries in GET /api/v1/system-info show a marketing name (e.g. "AMD Radeon RX 9070 XT (gfx1201)") instead of a raw numeric code.

The POST /v1/images/upscale endpoint requires the 'upscaling' model label; models without it return a 400 error — label your model 'upscaling' to use the endpoint.
OpenMOSS backend 'rocm' and 'rocm_bin' removed on Windows and Linux because ROCm builds run ~40x slower than Vulkan on the same hardware — use backend: vulkan or backend: cuda instead.
sd-cpp ROCm removed on Windows where the backend ran at CPU speed without any acceleration — switch to backend: vulkan or backend: auto.
AMD GPU entries in GET /api/v1/system-info now show a marketing name instead of a raw numeric code — parse the family field for the ISA code.
Editing user_models.json no longer takes effect without a restart of lemond; the restart requirement was always implied and is now documented.

Operating System	Downloads
Windows	lemonade.msi
Ubuntu 24.04+	Launchpad PPA
Debian 13 (x86_64)	lemonade-server_2026.40.0-debian13_amd64.deb
Debian 13 (ARM64)	lemonade-server_2026.40.0-debian13_arm64.deb
Fedora 43 (x86_64)	lemonade-server-2026.40.0-fc43.x86_64.rpm
Fedora 43 (ARM64)	lemonade-server-2026.40.0-fc43.aarch64.rpm
Fedora 44 (x86_64)	lemonade-server-2026.40.0-fc44.x86_64.rpm
Fedora 44 (ARM64)	lemonade-server-2026.40.0-fc44.aarch64.rpm
macOS	Lemonade-2026.40.0-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-2026.40.0-ubuntu-x64.tar.gz
Ubuntu arm64	lemonade-embeddable-2026.40.0-ubuntu-arm64.tar.gz
Windows x64	lemonade-embeddable-2026.40.0-windows-x64.zip
macOS arm64	lemonade-embeddable-2026.40.0-macos-arm64.tar.gz

Click to expand changelog

docs(embeddable): update packaging-time bundling limitations by @jeremyfowers in #3548
Report the AMD GPU marketing name on Linux by @apollo-2006 in #3601
fix(server): size streaming models against APU GTT pool, not VRAM carve-out (#3377) by @bong-water-water-bong in #3502
fix: dispatch image upscaling through backend capabilities by @atirna in #3511
feat(models): add Ornith 1.5 9B and 35B A3B to the catalogue by @noamsto in #3564
Don't run inference backend jobs on release branch builds by @jeremyfowers in #3604
fix(server): drop the lookup-miss registry reload by @jeremyfowers in #3613
Remove OpenMOSS ROCm support on all operating systems by @jeremyfowers in #3615
feat(server): parse and propagate cloud model token limits by @abn in #3568
refactor(mcp): make tool definitions declarative by @fl0rianr in #3619
fix: support mbedTLS 4 on Arch Linux by @fl0rianr in #3188
chore(backends): bump FastFlowLM to v1.0.6 by @zaneni6 in #3623
Add Junie by JetBrains support to lemonade launch by @mashan555 in #3621
Remove sd-cpp ROCm support on Windows by @jeremyfowers in #3616
docs: add stable platform install pages by @fl0rianr in #2923
ci(docs): check heading anchors by replacing markdown-link-check with lychee by @jeremyfowers in #3632
docs(models): name the custom model guide's sections after their topics by @jeremyfowers in #3640
ci: test PPA package upgrade path (config + model persistence) by @superm1 in #3396

Full Changelog: v2026.39.1...candidate-v2026.40.0

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