Skip to content
Drix10 Blog

NVIDIA Dynamo recipes add Korean open models, 2x B200 speed

, 8 items in AI Professionals and Community, 4 min read

In this digest (8 items)

NVIDIA Dynamo deployment recipes add Korea’s leading open models and claim up to 2x inference performance on B200 GPUs. The recipes cover four named models.

Key points

  • A.X K2: SKtelecom’s model included

  • K-EXAONE 2.0: LG_AI_Research’s model included

  • Solar Open 2: UpstageAI’s model included

  • Performance: up to 2x inference speed on NVIDIA B200 GPUs

Sources

Anthropic cost increase 8x over 2 years

Anthropic is now 8× as expensive to boost AI capabilities as it was 2 years ago, according to a quoted tweet. The post calls this a profound result indicating unsustainable fiscal scaling.

Key points

  • Cost per capability: 8× higher than 2 years ago.

  • Trend: Exponentially less capability per billion dollars spent.

Sources

AsyncVLA - real-time edge + remote inference for navigation

AsyncVLA enables real-time deployment of the OmniVLA navigation model using combined edge and remote inference. It achieves a 40% higher success rate than state-of-the-art baselines on vision-based navigation with up to 6 s latency. A presentation on the work will be given tomorrow at the poster session, Booth 142.

Key points

  • Real-time operation achieved by running the base VLA on a workstation and a lightweight Edge Adapter onboard the robot.

  • Quantitative results show a 40% increase in success rate compared to OmniVLA under delays up to 6 seconds.

Sources

NVIDIA AI Day Seoul Highlights Korea’s Advances in Sovereign, Agentic and Physical AI

NVIDIA AI Day Seoul gathers 2,000+ developers, startups and researchers in Seoul on Nov 9-10. NVIDIA announced new investments in Korean AI startups and released Dynamo-optimized deployment recipes for four leading Korean open models.

Key points

  • Investment announced in Korean startup Point2, with additional investments planned in Motif Technologies, RLWRLD and Upstage.

  • Dynamo-optimized deployment recipes released for A.X K2 (SK Telecom), K-EXAONE 2.0 (LG AI Research), Solar Open 2 (Upstage) and Motif-3 (Motif Technologies).

Sources

NVIDIA NemoClaw, OpenShell, K-EXAONE-2.5 API access announced

Hyundai Motor Group is exploring NVIDIA NemoClaw and OpenShell. AIMSafety is integrating OpenShell for secure AI agents. LG AI Research is opening API access to K-EXAONE-2.5 trained on NVIDIA Nemotron datasets.

Key points

  • Hyundai Motor Group exploring NVIDIA NemoClaw and OpenShell

  • AIMSafety integrating OpenShell for secure AI agent use

  • LG AI Research opening API access to K-EXAONE-2.5 trained on NVIDIA Nemotron datasets

Sources

Anthropic IPO roadshow begins week of November 9

Anthropic is expected to start its IPO roadshow the week of November 9, the earliest window available. SEC rules require at least 15 calendar days between a public filing and the start of investor meetings, and the company filed its.

Key points

  • IPO roadshow week of November 9, per CNBC.

  • SEC mandates 15-day gap between filing and investor meetings.

Sources

Flow Reversal Steering presented at CoRL on Tuesday

Flow Reversal Steering is a method for passing VLM “hints” to a diffusion VLA and will be presented at CoRL on Tuesday. Andy and Will will host a poster session at poster 39 on the 3rd floor.

Key points

  • Andy and Will will talk about Flow Reversal Steering at CoRL on Tuesday.

  • They will be at poster 39 at Lone Star (3rd floor).

Sources

RLT oral at CoRL2026, adds RL token to π-0.6

RLT will present an oral at CoRL2026 and describe adding an RL token to π-0.6 for quick fine-tuning. The talk is at 9 am and the poster is at 5 pm.

Key points

  • Oral O4 scheduled 9am-9: 30am

  • Poster #64 scheduled 5-6: 30pm

  • RL token output added to π-0.6 for a tiny actor and critic to learn quickly with RL

Sources