CONFIRMED: Subquadratic launched SubQ Snap on 12 October 2026. All benchmark results are COMPANY SELF-REPORTED. “First long-context decision model” is a COMPANY CLAIM.
Subquadratic has released SubQ Snap, which it describes as “a decision model for agentic workflows” [1]. The company says it is “the First Long-Context Decision Model Purpose-Built for Enterprise” (COMPANY CLAIM) [2].
A decision model does not write free text. It is given some context, a question and a fixed list of allowed answers, and it picks one, with a probability for each. AI agents make these small choices constantly: which file matters, which tool to use next, whether a test passed.
What Snap does
Subquadratic says Snap is built to take decisions “like choosing which files are relevant, which tool to use next, whether a test result passes, or which step needs a frontier model” [1]. Its argument is that “today’s decision models stay at 32K tokens and decide without most of what they need to see” [1].
According to the company, Snap:
- “makes each decision with up to 384K tokens of text and images” [1];
- “processes requests up to 100K tokens in under a second” [1];
- accepts “screenshots, charts, diagrams, and scanned pages” alongside documents and logs [1].
A token is a small chunk of text, roughly three-quarters of a word. 384K tokens is enough for a large codebase or a long contract with its exhibits.
Business Wire’s release says the context window is “12x larger than Jev”, TypeSafe AI’s decision model (COMPANY CLAIM) [2].
How it works
Snap runs on Subquadratic Sparse Attention (SSA). The company says SSA “reduces attention compute cost from quadratic to linear complexity” [1]. In a standard model, doubling the input roughly quadruples the attention work; Subquadratic says its design avoids that. Per the Business Wire release, “In research testing at 12 million tokens, the architecture cut attention compute by almost 1,000x compared with other frontier models” (COMPANY CLAIM) [2].
What Subquadratic says about performance (COMPANY SELF-REPORTED)
All results below come from Subquadratic’s own tests against Jev:
| Benchmark | SubQ Snap | Jev |
|---|---|---|
| Decision Index 0.2.1 (short requests, under about 34K tokens) | 61.7 | 57.9 |
| Decision LCR (long multi-document inputs) | 74.0 | 48.0 (chunked) |
| 16 enterprise Decision Index benchmarks (average) | 67.6 in the text, 67.54 in the table | 61.6 |
| Bespoke Labs’ Nimble (13 datasets) | 82.9% | 76.0% |
The company says Snap “leads on 13 of 16” enterprise benchmarks, with the biggest gains “in phishing detection, code execution prediction, financial sentiment, customer-support routing, and contract review” [1]. It trails Jev on three: API-Bank, When2Call and ACOS [1].
Subquadratic notes that “Decision LCR scores can’t be compared with published AA-LCR results, which grade written answers rather than decisions” [1].
A benchmark is a standard set of test tasks. When a company picks and runs its own, the results are not independent.
Price and availability
- “SubQ Snap is priced at $0.05 per million input tokens, with no charge for output tokens” [1].
- “SubQ Snap is available in Public Preview on a first come, first served basis while capacity is limited. Once capacity is reached, new users may be added to a waitlist” [1].
- “SubQ 1.1 Small, our Language Model, is in Early Access with selected design partners” [1].
What the CEO says
“Decision models get exciting when they can take real work away from LLMs, but the context window size of today’s decision models are simply too small or slow to handle real enterprise work,” said Alex Whedon, co-founder and CEO of Subquadratic [2].
Sources
- Subquadratic, “Introducing SubQ Snap: A Long-Context Decision Model for Enterprise Agents”, 12 October 2026. https://subq.ai/introducing-subq-snap
- Business Wire (via FinancialContent), “Subquadratic Launches SubQ Snap, the First Long-Context Decision Model Purpose-Built for Enterprise”, 12 October 2026. https://www.financialcontent.com/article/bizwire-2026-10-12-subquadratic-launches-subq-snap-the-first-long-context-decision-model-purpose-built-for-enterprise
Related stories
- TypeSafe AI, Maker of Jev, Raises $870 Million at a $7.5 Billion Valuation
- OpenAI’s Decisions API and Perplexity’s Open Decider: AI That Answers With a Probability, Not an Essay
- RAG vs Jev+RAG Explained: What the Judgment Gate Adds
- Small Open Models That Answer in One Pass: Liquid AI’s d1 and Perplexity’s pplx-embed-v2-late
