v1.0.0-beta.234 v0.30.0
high risk
ruvnet/RuViewSep 25, 2026 → Sep 25, 20260 days between releases
What's New
Shift from billing/entitlements APIs to a unified AI serving framework and vLLM rebase
Breaking Changes
- `diffusion_batch_size` has been removed; use `max_num_seqs` instead.
New Features
- Unified full-duplex serving framework (DuplexOmni, DuplexOrchestrator).
- Native cross-stage KV and multimodal payload transfer.
- MAGI-2 Preview text/image-to-video generation.
- Interactive world-model and streaming video generation.
- Support for MiniMax H3 on Nvidia Blackwell and Ascend NPU 950.
Improvements
- Improved batching and memory efficiency.
- Extended scheduler-managed diffusion KV into cross-stage transfer.
- Added KV prefetch and HunyuanImage3 cross-request paged prefix caching.
- Native Mooncake AR-to-DiT handoff.
- Separated duplex orchestration from turn-based execution.
Upgrade Notes
Replace `diffusion_batch_size` with `max_num_seqs` in configuration and be aware of the ledger refactor if upgrading from the beta.
The Two Releases
v1.0.0-beta.234
v1.0.0-beta.234
Sep 25, 2026· 27 days ago
Adds v3 APIs for customer entitlements, grants, and value, along with various billing fixes including invoice deletion idempotency and negative metered usage handling.
Full release notesv0.30.0
v0.30.0
Sep 25, 2026· 26 days ago
A major release featuring 504 changes including a unified full-duplex serving framework, native cross-stage KV transfer, and expanded model support. Rebased onto vLLM 0.30.0.
Full release notes