All Articles · SURL AI News

FLUX 3 Action releases weights, jointly denoising video and actions

The model offers DROID and SO-101 weights and fine-tuning workflows, with a success rate of about 42.9% on the public RoboLab leaderboard. Applications must still manage the control loop, and the planning horizon must not be mistaken for inference latency.

2026-09-23

NVIDIA unveils eight-speaker diarization model with minimum recommended buffering latency of 320 milliseconds

Nemotron 3 Diarization uses 100 million parameters to process streaming and recorded audio, retaining a speaker cache to keep labels consistent across segments. The latency figures cover only audio buffering; actual deployments still need to measure speech recognition, transmission, and post-processing costs.

2026-09-23

GitHub Repo

Npunlock validates custom C kernels on Intel NPU, runs independent mixed-precision branches in one graph

The project uses Intel’s driver to compile computation graphs, then replaces SHAVE kernels to run custom C code on the NPU. New tests cover independent branches at two precisions, but arbitrary mixed-precision pipelines are not yet supported.

2026-09-23

GitHub Repo

PyTorch 2.14 in-place transpose errors reported, implicating CUDA compiler fusion

An RTX 4070 reproducer shows incorrect results for certain tensor sizes after torch.compile. The report points to loop reordering and read/write dependencies; a fix and the impact across environments remain unconfirmed.

2026-09-23

AI 研究

Extending Recursive Model Inference to 512 Steps Raises Cumulative Sudoku Solve Rates

By extending computation with the same weights, researchers observed that some initially unsuccessful puzzles were solved later. The findings offer clues for studying stopping criteria, but do not yet establish cost-effectiveness in deployed services.

2026-09-23

AI 研究

GPU Topology Study Adjusts Memory Allocation, Measures 14.3% Gain in H200 Decode Throughput

The study incorporates compute unit placement and memory partitions into inference configurations, improving decode throughput under a specific Qwen3-8B workload. The prototype requires driver modifications, and the results await code release and reproduction across environments.

2026-09-23

GitHub Repo

Langflow 1.12.3 fixes cross-user knowledge base isolation; legacy data migration requires manual review

The new release separates OpenSearch and Chroma Cloud knowledge base storage by owner to prevent resources with identical names from sharing storage. Legacy data with uncertain ownership is retained, but knowledge bases may be redirected to empty storage.

2026-09-23

GitHub Repo

vLLM 0.30.0 adds resident GPU weight caching for reuse across engine restarts

Fast Start reuses processed weights through CUDA IPC, reducing disk loading during engine restarts. Zero-copy mode is incompatible with sleep mode, and deployers must also validate parallelism configurations and cache fallback behavior.

2026-09-22

Claude Opus 5.5 lowers token prices and introduces multiple breaking changes to the agent API

The new version reduces input, output, and cache read prices and makes adaptive thinking always active. Forced tool calls, edits to conversation history, and streamed progress displays all require renewed validation.

2026-09-22

GitHub Repo

vLLM publishes Qwen3.8-2.4T disaggregated inference benchmarks, reaching about 5,000 total tokens per second per GPU

The team benchmarked separate prefill and decode configurations on GB300 NVL72, providing version details and a configuration repository. Peak throughput and roughly 180 generated tokens per second per user come from different configurations and still require validation under real workloads.

2026-09-22

AI 研究

Shared Inference Cost Study Publishes H100 Results, Showing How Metering Changes Tenants’ Cost Shares

unalloc consolidates cluster and model bills. The study found that different metering methods shifted the same tenant’s share of costs by roughly 12–14 percentage points. Hardware validation consisted of one short run per workload, so the results do not establish a definitive allocation.

2026-09-22

GitHub Repo

Hermes patches Claude CLI plugin so tools with no arguments no longer trigger fallback provider switches

Nous Research updates the DirectSDK plugin’s pinned commit to fix issues with tools that take no arguments, thinking settings, and interface compatibility. The version remains 0.3.0, so deployers must check the commit hash and actual tool execution flow.

2026-09-22