Hugging Face shipped transformers v5.18.0 on 2026-09-30 (PyPI 5.18.0): NVIDIA Nemotron 3 Diarization (streaming/offline, up to 8 speakers, 80 ms–30.4 s buffers), NemotronH Omni (joint text/image/video/audio), NAVER HyperCLOVAX Vision V2 (built-in <think> CoT), and GTE bidirectional encoders (RoPE + gated MLP; gte / snowflake-arctic-embed family). Official model_doc pages are live.

Key Takeaways

  • ✓Release v5.18.0 at 2026-09-30T16:46:27Z; PyPI ~2026-09-30T16:45:57Z
  • ✓Nemotron 3 Diarization: streaming+offline, ≤8 speakers, 80 ms–30.4 s buffers, 10 ms frame multiples (#49056)
  • ✓NemotronH Omni: NemotronH + RADIO + optional Parakeet for joint text/image/video/audio (#46509)
  • ✓HyperCLOVAX Vision V2 (NAVER) + GTE encoders (RoPE/gated MLP; gte / snowflake-arctic-embed family)
  • ✓pip install -U transformers==5.18.0; docs for diarization/omni/gte linked above
Hugging Face Transformers 5.18.0: Nemotron 3 streaming diarization; NemotronH Omni; HyperCLOVAX Vision V2; GTE
🖼️Official Media
Click to view high-res
⚡

Key Decision Metrics at a Glance

TL;DR Verdict
🏆 Benchmark RankRank
No model benchmark in this story · see current rankings
Full Leaderboard→
💰 Cost & PlansCompare
API & Subscription
Compare Plans↗
🛠️ Ecosystem & IDERules
Transformers · Huggingface · Nemotron
Setup IDE Rules↗
🧭

Turn your technical choice into a development budget

Compare 40 dev plans & simulate token costs vs $20/mo subscriptions

⚔️ Compare Model→💰 Compare plans & limits↗
🔗

Project Links & Resources

Direct Access
Direct access to official project resources and documentation
🐙
transformers v5.18.0 release
github.com/huggingface/transformers/releases/tag/v5.18.0
↗
📖
transformers 5.18.0 on PyPI
pypi.org/project/transformers/5.18.0/
↗
📖
Nemotron 3 Diarization docs
huggingface.co/docs/transformers/main/en/model_doc/nemotron3_diarization
↗
📖
NemotronH Omni docs
huggingface.co/docs/transformers/main/en/model_doc/nemotron_h_omni
↗
📖
HyperCLOVAX Vision V2 docs
huggingface.co/docs/transformers/main/en/model_doc/hyperclovax_vision_v2
↗
📖
GTE docs
huggingface.co/docs/transformers/main/en/model_doc/gte
↗
🌐
Paper 2507.18446 (AOSC/Sortformer)
huggingface.co/papers/2507.18446
↗
🐙
#49056
github.com/huggingface/transformers/pull/49056
↗
🐙
#46509
github.com/huggingface/transformers/pull/46509
↗
🚀
2409.06656
huggingface.co/papers/2409.06656
↗
🚀
2407.19669
huggingface.co/papers/2407.19669
↗
🔬

In-Depth Technical Analysis

Background

Streaming diarization and joint text/image/video/audio stacks often lived outside Transformers. v5.18.0 pulls Nemotron 3 Diarization, NemotronH Omni, HyperCLOVAX Vision V2, and GTE into the main library with docs.

What shipped

[v5.18.0](https://github.com/huggingface/transformers/releases/tag/v5.18.0): Nemotron 3 Diarization (≤8 speakers, 80 ms–30.4 s buffers) [#49056]; NemotronH Omni (RADIO + optional Parakeet) [#46509]; HyperCLOVAX Vision V2 with <think>; GTE (RoPE + gated MLP).

Benchmarks

No single SWE-bench table in the notes. Papers: 2507.18446 / 2409.06656 (diarization), 2407.19669 (GTE). Anchors: GitHub 2026-09-30T16:46:27Z, PyPI 5.18.0. Measure your own streaming latency and VRAM.

Get started

pip install -U transformers==5.18.0. Docs: https://huggingface.co/docs/transformers/main/en/model_doc/nemotron3_diarization

Action HubReady to adopt this in production?

Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.

⚔️ Launch VS Compare→💳 View Plans↗
Up NextNVIDIA open-sources Nemotron 3.5 Transcribe, a 30B MoE speech model with about 4B active parameters that handles transcription in 40 languages, speaker attribution and translation, and tops open models on the Open ASR English leaderboard
Read Now→