v1.8.0

🚀 Added

🦋 Gemini Argon cuts out real shapes

gemini_vlm@v6 now runs instance-segmentation on gemini-4-argon. One object can be several detached pieces with holes in them, a donut or a fence seen through foliage, and still come back as a single RLE mask with a box derived from that mask. Other Gemini models refuse the task up front instead of failing halfway (@SkalskiP, #3189).

🎨 Meta block meets Muse Wonder

meta_vlm@v3 gains Muse Wonder with reasoning levels from minimal to xhigh, decoding boxes the same way as the Spark models. Heads-up: it is now the default, so workflows that omit model_version call Muse Wonder instead of Muse Spark 1.3 (@SkalskiP , #3186).

🎟️ Pick your OpenAI lane

openai@v7 takes an optional service_tier: flex for cheap overnight batches, priority or ultrafast when someone is staring at the screen. It works with your own key and with Roboflow-managed rf_key: keys. Unset, nothing changes (@iurisilvio, #3194).

🧬 RF-DETR NAS learns segmentation

NAS-searched RF-DETR semantic segmentation models (rfdetr-nas-sem) now load on the server instead of hitting NoModelPackagesAvailableError. Bonus fix: freshly compiled NAS TensorRT packages for detection and instance segmentation were silently rejected, so the server fell back to ONNX. They load as TRT now (@leeclemnet, #3187).

🪶 CPU images lose 3 GB of luggage

The CPU images installed CUDA PyTorch they could never use: about 15 nvidia-* packages plus Triton. They now install CPU-only wheels. Compressed size drops from 5.1 GB to 2.2 GB (-58%) for inference CPU and from 5.3 GB to 2.3 GB for edge CPU (@ColePBryan, #3137).

🔧 Fixed

• Custom Python blocks stop eating RAM — since 1.6.1 every HTTP request recompiled workflows with custom Python blocks, leaking about 7 MB per request until dedicated deployments OOM'd. Compiled workflows are cached again, and evicted ones no longer stay pinned by global metadata caches (@grzegorz-roboflow, #3196; @dkosowski87, #3197).
• Remote model calls keep their workflow attribution — models called from a tagged workflow on a remote server were billed to "Deployment"; source and source_info now travel with v0, v1 and core-model requests (@joaomarcoscrs, #3177).
• Fine-tuned Florence-2-large answers again — LoRA fine-tunes returned empty strings because decoding repeated <s> until the token limit; that one bigram is now banned (@imbgar-roboflow, #3115).
• Anomaly maps reach v0 clients — include_anomaly_map was dropped on the legacy v0 route, the default for hosted endpoints (@probicheaux, #3202).
• Custom Python executors survive a full region — Modal executors were pinned to aws/us-east-1 and could not reschedule when that region ran out of capacity; US executors may now land in any US region. EU stays put for data residency (@rafel-roboflow, #3206).

⚙️ Execution Engine v1.16.1

Stays at 1.16.1. Custom Python compilation caching is restored and the platform blocks are core again; existing workflows need no migration. Details in workflows/CHANGELOG.md.

🚧 Maintenance

• Versions — inference 1.8.0, inference-models 0.39.4, roboflow-workflows 0.2.9 (@PawelPeczek-Roboflow, #3190; @MTaraszewski, #3207, #3208).
• Faster CI — parallel unit tests, per-workflow pip caches, the torch_models CPU suite sharded from about 20 min to about 7 min, docs-only changes skip code tests, duplicate Modal runs dropped and smaller runners where they fit (@alexnorell, #3164, #3165, #3166, #3167, #3168, #3169, #3170).
• Test hygiene — unit tests no longer report usage to the live API, and two tests stop tripping over background threads (@alexnorell, #3171, #3172); refreshed TensorRT expected predictions after the tensor fixes (@grzegorz-roboflow, #3184, #3188).
• CI images and hosted E2E — CPU and GPU images reused across server test modes, plus four-shard hosted E2E acceptance (@MTaraszewski, #3200, #3199); a cleaner maintainer handoff comment (@grzegorz-roboflow, #3198).

Full Changelog: v1.7.4...v1.8.0

Windows installer update: 1.8.0-post1 (2026-10-20)

Updated native Windows installers repair missing torchvision native operators and explicitly select matching CPU/CUDA PyTorch wheels. The bundled Inference version remains 1.8.0. These installers were built from commit 80d7ecd50597a550365a7bf4856e463a17d99960, based on the v1.8.0 tag with the Windows packaging fixes from PR #3214.

• Download the 1.8.0-post1 CPU installer (CI build)
• Download the 1.8.0-post1 GPU installer (CI build)

GPU compatibility: The native Windows GPU installer uses PyTorch CUDA 12.6 builds. Blackwell GPUs, including GeForce RTX 50-series, are unsupported by this installer. Use the CPU installer for CPU execution on these systems. This limitation concerns the native Windows installer; Docker images have separate compatibility requirements. An ONNX backend alone does not establish a supported Blackwell inference path, because preprocessing and postprocessing can also use PyTorch CUDA.

SHA-256 checksums of the installer executables:

fb5a71536bc79f022e0d0642fc32c767f318e8d4b67b190bd51cf8ab669e72f2  inference-1.8.0-post1-cpu-installer.exe
d3a2502309de11fbdb72a3437ec3f0ad2326cb8db672775cac6e30cbb75d3d44  inference-1.8.0-post1-gpu-installer.exe