SSyncDev

SubQ Snap 1.0

SyncDev model directory · Profile updated Nov 18, 2026

Compare with another model

SubQ Snap 1.0 was Subquadratic's first long-context decision model: 384K tokens of text and images, linear attention, $0.05 per million input tokens.

Key specifications

Specification score
31.8
Context window
384K
Input / 1M tokens
From $0.05
License
Proprietary

The specification score summarizes recorded features and limits. It is not a benchmark result.

Pricing conditions: Launch price $0.05 per 1M input tokens, output free; users are now auto-upgraded to Snap 1.2. Confirm on console.subq.ai.

Inputs and outputs

Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.

Accepts

  • Text input
  • Image input

Produces

  • Text output

Capabilities

Vision

What is SubQ Snap 1.0?

What if a model never wrote a single output token? SubQ Snap 1.0, released by Subquadratic on 12 October 2026, was built on that premise. It reads a long state of mixed text and images and maps it, through a decision head, to a probability over a fixed set of allowed answers. No autoregressive generation, so output was free.

That makes it a different tool from a chat model. You use it to route a ticket, approve or reject a claim, or pick one of five categories, not to draft a reply.

The architecture bet

Subquadratic's core claim is Subquadratic Sparse Attention, which it describes as cutting attention cost from quadratic to linear in input length. The practical promise was that deciding over 100,000 tokens takes about one second, and that a 384K-token state costs the same per token as a short one.

Item 1.0 at launch
Released 12 October 2026
Context Up to 384K tokens, text and images
Input price $0.05 per 1M tokens
Output Free
Access Public preview, waitlist when capacity ran out

Launch numbers, and the gaps in them

Subquadratic compared Snap mainly with TypeSafe's Jev, the obvious incumbent in this niche. Its own figures:

  • Decision Index 0.2.1: 61.7, against 57.9 for Jev
  • average across 16 enterprise benchmarks: 67.6 against 61.6, ahead on 13
  • Decision LCR: 74.0 in one shot, against 48.0 for Jev with chunking
  • Bespoke Nimble: 82.9% against 76.0%

Two things in the launch material do not add up. Decision RULER scores were quoted out to 1M tokens, which is beyond the stated 384K window, while the accompanying table only ran to 256K. And the later 1.2 announcement describes Snap models as covering 256K tokens. If you need a firm limit, test it yourself with a long synthetic state before relying on either figure.

Where 1.0 stands now

It lasted ten days as the current release. SubQ Snap 1.1 followed on 22 October with better calibration and a lower price, and SubQ Snap 1.2 arrived on 11 November. Subquadratic says users on earlier versions are moved to 1.2 automatically at no cost, so in practice you cannot pin this version for new work. Its page is mainly useful as a baseline when reading later benchmark comparisons, many of which quote 1.0 figures.

If you are evaluating the family

Build a set of 200 to 500 historical decisions with known outcomes, include several longer than 100K tokens, and run them through the current version. Check accuracy, then check that a 0.9 probability is right about nine times in ten. Compare against Jev on the same set. Other decision endpoints are collected under structured decisions.

How to evaluate SubQ Snap 1.0

Compare the model on the work you need to complete.

Tasks and inputs

SubQ Snap 1.0 is listed for Text classification, Structured Decisions. It accepts text, image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.

Capacity and cost

Documented limits: 384K context. The listed route has starting rates of $0.05 input per million tokens. Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.

Choosing a deployment

Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.

Frequently asked questions

Subquadratic released it on 12 October 2026 in public preview.

These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.

Microsoft

Microsoft-Decision-1.1

Microsoft-Decision-1.1 is a decision model for routing, triage and classification, returning calibrated probabilities. GA in Foundry from November 2026.

— context

Liquid AI

d1.2

Liquid AI d1.2 is a hosted omnimodal decision model for text, image and audio questions with calibrated answers, at $0.02 per million input tokens.

65.54K context

Liquid AI

d1.2-ultra

Liquid AI d1.2-ultra is the top hosted decision model in Liquid's line, scoring 75 on the AA Decision Index, with audio support at $0.042 per 1M input.

65.54K context

Google

Gemini Decision 1

Gemini Decision 1 powers Google's Decisions API, returning calibrated yes/no, choice and rubric answers instead of text, at $0.10 per 1M input tokens.

262.14K context

subquadratic

SubQ Snap 1.2

SubQ Snap 1.2 is Subquadratic's long-context decision model on linear sparse attention, tested to 256K tokens, priced at $0.04 per 1M input tokens.

— context

Perplexity

pplx-decider-flash-v1.3

pplx-decider-flash-v1.3 is Perplexity's small, half-price decision model, distilled from the 27B decider, for high-volume classification and span extraction.

— context