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, https://github.com/roboflow/inference/pull/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 , https://github.com/roboflow/inference/pull/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, https://github.com/roboflow/inference/pull/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, https://github.com/roboflow/inference/pull/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, https://github.com/roboflow/inference/pull/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, https://github.com/roboflow/inference/pull/3196; @dkosowski87, https://github.com/roboflow/inference/pull/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, https://github.com/roboflow/inference/pull/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, https://github.com/roboflow/inference/pull/3115).
- **Anomaly maps reach v0 clients** — `include_anomaly_map` was dropped on the legacy v0 route, the default for hosted endpoints (@probicheaux, https://github.com/roboflow/inference/pull/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, https://github.com/roboflow/inference/pull/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`](https://github.com/roboflow/inference/blob/v1.8.0/workflows/CHANGELOG.md).

# 🚧 Maintenance

- **Versions** — `inference` `1.8.0`, `inference-models` `0.39.4`, `roboflow-workflows` `0.2.9` (@PawelPeczek-Roboflow, https://github.com/roboflow/inference/pull/3190; @MTaraszewski, https://github.com/roboflow/inference/pull/3207, https://github.com/roboflow/inference/pull/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, https://github.com/roboflow/inference/pull/3164, https://github.com/roboflow/inference/pull/3165, https://github.com/roboflow/inference/pull/3166, https://github.com/roboflow/inference/pull/3167, https://github.com/roboflow/inference/pull/3168, https://github.com/roboflow/inference/pull/3169, https://github.com/roboflow/inference/pull/3170).
- **Test hygiene** — unit tests no longer report usage to the live API, and two tests stop tripping over background threads (@alexnorell, https://github.com/roboflow/inference/pull/3171, https://github.com/roboflow/inference/pull/3172); refreshed TensorRT expected predictions after the tensor fixes (@grzegorz-roboflow, https://github.com/roboflow/inference/pull/3184, https://github.com/roboflow/inference/pull/3188).
- **CI images and hosted E2E** — CPU and GPU images reused across server test modes, plus four-shard hosted E2E acceptance (@MTaraszewski, https://github.com/roboflow/inference/pull/3200, https://github.com/roboflow/inference/pull/3199); a cleaner maintainer handoff comment (@grzegorz-roboflow, https://github.com/roboflow/inference/pull/3198).

**Full Changelog**: https://github.com/roboflow/inference/compare/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](https://github.com/roboflow/inference/pull/3214).

- [Download the 1.8.0-post1 CPU installer](https://github.com/roboflow/inference/releases/download/v1.8.0/inference-1.8.0-post1-cpu-installer.exe) ([CI build](https://github.com/roboflow/inference/actions/runs/39422374403))
- [Download the 1.8.0-post1 GPU installer](https://github.com/roboflow/inference/releases/download/v1.8.0/inference-1.8.0-post1-gpu-installer.exe) ([CI build](https://github.com/roboflow/inference/actions/runs/39422389244))

**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:

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