Tasks and inputs
SubQ Snap 1.2 is listed for Text classification, Structured Decisions. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 17, 2026
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.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: $0.04 per 1M input tokens, no output charge (unchanged from Snap 1.1). Indicative; confirm on your own account.
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
Produces
Reorder the answer options and a classifier should give the same answer. Many do not. On Subquadratic's 600-question test, its previous Snap 1.1 changed its top answer 16.0% of the time when options were rotated. SubQ Snap 1.2, released on 11 November 2026, brings that to 0.0%.
That is the most practical change in this release, and it is a reason to retest any pipeline built on 1.1.
Snap runs on Subquadratic Sparse Attention, a linear-attention algorithm, so per-token compute stays flat as input grows. That makes it a decision model you can point at a whole contract, a long ticket history or a filing, rather than a snippet. Snap 1.2 was benchmarked from 1K to 256K tokens; on a key-lookup abstention test it holds about 98% or better through 256K.
| Snap 1.1 | Snap 1.2 | |
|---|---|---|
| Price | $0.04 / 1M input | $0.04 / 1M input |
| Top-answer change under option reordering | 16.0% | 0.0% |
| HF Decision Index v0.3 (open partition) | 66.43 | 69.05 |
| Decision LCR v1.1 | 63.2 | 68.4 |
| Upgrade | n/a | Automatic for existing users |
New post-training stages covered long-context recall, state tracking and multi-hop reasoning, plus synthetic enterprise documents with known answers. Subquadratic also reports a tie for first, at 73, on an independent decision index. All publisher-reported.
Subquadratic is unusually open about its gaps. Jev 1.14 leads on a multi-round coreference test between 16K and 128K tokens. The reasoning split of its abstention test is hard for every model. And on its 16-task enterprise suite, the text says 1.2 significantly improves on 7 of 16 subtasks while a figure caption says all sixteen differences are positive; the two statements are not reconciled. Read the 7-of-16 claim as the conservative one.
It also notes that d1-ultra's 32K window is too short for its long-document tests, which is a fair point about fit rather than quality.
No API model identifier or maximum context is printed on the 1.2 release. Access is through Subquadratic's console. Earlier versions (the original Snap from mid-October and 1.1, a public preview with a 384K window listed on 22 October) are described only in third-party catalogues, so we do not rely on those details. Since existing users are moved to 1.2 automatically, there may be no way to pin 1.1; check before relying on reproducibility.
Assemble 100 long documents with labelled answers, at the lengths you really see. Ask each question twice with the options in different orders, and record both accuracy and agreement between runs. Then repeat at half and double the length to see where performance bends. Other models for this kind of work are under structured decisions, including Jev.
Compare the model on the work you need to complete.
SubQ Snap 1.2 is listed for Text classification, Structured Decisions. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
The listed route has starting rates of $0.04 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.
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.
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 is a decision model for routing, triage and classification, returning calibrated probabilities. GA in Foundry from November 2026.
Liquid AI
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.
Liquid AI
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.
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.
Perplexity
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.
Perplexity
pplx-decider-v1.3 is Perplexity's 27B-class decision model: it returns probabilities for choice, yes/no, score and new span-locate questions, not text.