Search papers, labs, and topics across Lattice.
K\mathcal{F}=\{f_{j}\}_{j=1}^{K} includes both closed-source and open-weight MLLMs with heterogeneous architectures, capacities, and cost profiles: • Commercial MLLMs, including GPT-5 series [1], Gemini 2.5 series [25], and Claude models, which offer strong general-purpose multimodal reasoning at higher monetary cost. • Open-weight MLLMs, spanning, Plus MMStar RealWorldQA Method
1
0
3
6
Forget hand-tuning rollout budgets: $V_{0.5}$ dynamically allocates compute to sparse RL rollouts based on a real-time statistical test of a generalist value model's prior, slashing variance and boosting performance.
