Subquadratic Launches SubQ Snap, a Long-Context Decision Model with 384K Token Context

Startup Subquadratic released SubQ Snap on October 13, positioning it as the first long-context "decision model" built for enterprise agent workflows. Key details come from a roundup of reposted threads: the model supports a 384K-token multimodal context, officially claimed to be 12x larger than competitor Jev; its biggest technical selling point is that compute per token stays constant as context grows (the subquadratic property). On pricing, it costs just $0.05 (5 cents) per million input tokens.

Confirmed

• Product named SubQ Snap, released by Subquadratic, positioned as an enterprise long-context "decision model"
• 384K-token multimodal context, officially claimed to be 12x larger than Jev
• Compute per token remains constant as context grows
• Priced at $0.05 per million input tokens
• Decision Index v0.2.1 score of 61.7 (versus Jev; the exact comparison scores weren't fully shown in the reposts)
• The model reads the entire conversation each turn (messages, PDFs, images, whole GitHub repos) and can directly decide whether the UI should render a chart or text, claiming decisions in 2-3 seconds
• Official demo reportedly got the Chrome dinosaur game working

Why it matters

If the "constant compute per token" claim holds, SubQ Snap could dramatically cut inference costs for long-context workloads, making large-scale deployment of enterprise agents over full codebases and lengthy documents far more economical. Its "decision model" positioning also sets it apart from general-purpose generative models, targeting routing and interface decisions within agent workflows. That said, all posts in this cluster are reposts; original sources for the comparison data (such as Jev's Decision Index score) and technical details should be verified against the official release.

2026-10-13 ~ 2026-10-13 · 7 related posts

• SubQ Snap launches: 384K-context decision model at $0.05 per million tokens — ryanallred · 2026-10-13
• SubQ Snap launches: 384K-context decision model at $0.05/M tokens, beat a Chrome dinosaur high score — ryanallred · 2026-10-13
• SubQ Snap launches: 384K-token decision model at $0.05/M, beats Jev on Decision Index — Med1_Ai · 2026-10-13
• Snap's new decision model handles 384K tokens per call at $0.05/M input, free output — thisdudelikesAI · 2026-10-13

3 near-duplicate retellings: ryanallred · ryanallred · ryanallred