LMSYS Releases Inkling-small: 12B Active Parameters Lowers Multimodal RL Barrier
simonguozirui · x · 2026-07-31
LMSYS has released the new MoE model Inkling-small. With 276B total parameters but only 12B active, it is considered a sweet spot for multimodal Reinforcement Learning (RL).
Performance & Inference: Using the SGLang framework, the model achieves 648 tok/s on 8x B200 GPUs. Its capability matches the larger version and even wins on some benchmarks.
Industry Significance: The 12B active parameter count means small teams can actually run multimodal RL, not just read papers about it. Combined with the Miles framework, developers can directly turn multimodal data into real capability gains, marking an industry shift from 'collecting data hoping labs train on it' to 'customizing models autonomously'.
Related event: Inkling-Small: 276B params, 12B active, open-source tops ARC(48 posts)→
More from Models
- Reflection's 501B MoE Beam hits 463 tok/s on 8x GB200 with NVFP4 at day 0 — zhyncs42 · 2026-10-16
- JevBench v1.7.46 leaderboard: new d1-ultra tops 32 API systems at 76.1 composite — airesearch12 · 2026-10-16
- LMArena launches Agent Arena and Code Arena leaderboards for model comparison — arena · 2026-10-16
- Third-party tests find NVIDIA Nemotron-3.5 Bolt underperforms nano on pandas/numpy and decision-making tasks — norenEnmotalen · 2026-10-16
- Reflection open-sources Beam under Apache 2.0 with 40+ partner ecosystem — bhutanisanyam1 · 2026-10-16
- One Prompt, 629 Lines: IQuest-Q1 Builds a Deep-Sea Submarine Game in 2 Minutes — Div_pradeep · 2026-10-16