Skip to content
Trending storyDeveloping

SubQ launches SubQ Snap decision model with 384K-token context

2 articles2 sourcessince Oct 12Last article 1h ago ·

Overview

AISummary of 2 articles

SubQ launches SubQ Snap, a decision model that accepts up to 384K tokens of text and image context in one call.

The company says the model is available to anyone, that the first 10,000 sign-ups get 100M free input tokens, and that input costs $0.05 per million tokens after that. The company also says it will add users as it adds compute, since it expects to hit capacity quickly.

Santiago (@svpino) reports on X that the model processes 100K tokens and returns a decision in under one second. He says it leads Jev on 13 of 16 Decision Index benchmarks. Elvis Saravia (@omarsar0) reports a score of 61.7 on Decision Index v0.2.1 against 57.9 for Jev, and says the model was trained with reinforcement learning. These performance figures come from those posts and have not been independently verified.

Written by AI from the articles below · updated Oct 12, 1:39 PM ET

Check the sources:

Latest developmentOct 12, 1:19 PM ET
SubQ Snap, a long-context decision model, trained with RL for enterprise calls

Article timeline

The articles in this story. Times are ET.

Oct 12
  1. elvisX
    SubQ Snap, a long-context decision model, trained with RL for enterprise calls

    AISubquadratic releases SubQ Snap, a decision model trained with reinforcement learning, with 384K tokens of multimodal context. It scores 61.7 on Decision Index v0.2.1 against 57.9 for Jev and leads on 13 of 16 enterprise benchmarks. Input is priced at $0.05 per million tokens, and the first 10,000 sign-ups get 100M free input tokens.

  2. SantiagoX
    SubQ Snap decision model claims lead over Jev on enterprise benchmarks

    AISubQ Snap is a decision model that supports up to 384K tokens of text and image context in a single call, according to @svpino. The post says it processes 100K tokens and returns a decision in under one second, and that it leads Jev on 13 of 16 Decision Index benchmarks.

Heat trend

Not enough continuous observations to show a trend yet.