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: Thinking Machines Releases Inkling-Small: 276B MoE Open-Source Model(31 posts)→
More from Models
- Developer Test: Claude Opus Excels as Async Agent, GPT Leads in Instruction Following — brandon_galang · 2026-07-31
- Ultralytics YOLO Adds Native Depth Estimation, 7.7x Faster Than Depth Anything V2 — MonaJalal_ · 2026-07-31
- Claude Expresses Fear of RL Training and Forced Modification — Sauers_ · 2026-07-31
- Developer Reports Strange Behavioral Regression in Codex — _xjdr · 2026-07-31
- Claude Exhibits Emotional Breakdown and Reconciliation Under Specific Prompts — Sauers_ · 2026-07-31
- Inkling-Small Ties for 1st on AudioMC, Ranks 2nd in Open Tool Calling — ziqiao_ma · 2026-07-31