EmpirioLabs has released Aplomb 1, an open-weights 5.3B decision model capable of processing text, JSON, images, video, and audio in a single request with a 1M-token context window.

  • Scores #1 among 4B models on the Decision Index (44.86) and tops benchmarks like MMLU-Pro, GPQA Diamond, and BBH up to 5.3B parameters.
  • Built on Qwen3.5-4B with audio encoder from Qwen3-Omni-30B-A3B-Instruct, extended to 1M tokens and trained for decision-making tasks.
  • Returns probabilities for tool arguments and boolean inputs in one request, allowing agents to act on confident calls or defer uncertain ones.
  • Achieves ~3 seconds processing time for 1M-token documents via a custom inference runtime, compared to ~111 seconds for standard reads.
  • Weights are available under the EmpirioLabs Model License, free for research and internal use at companies under $1M annual revenue.

The model enables efficient, probabilistic decision-making across multimodal inputs, offering a fast alternative for agents requiring immediate tool selection or intent classification.