Subquadratic
Verified Oct 15, 2026 ActiveLong-context LLM (SubQ) built to reason over million-token documents without chunking, in early access.
Listed on EffectHub ·
Pricing
- Pricing model
- Contact sales
Not yet priced; access is by API request during early access.
Use tool ↗About Subquadratic
Subquadratic Inc. is a research company, founded by engineers and researchers with backgrounds at Meta, Google, DoorDash, Amazon, and other firms, building a model called SubQ that targets multi-million-token reasoning. Instead of splitting a large codebase, contract archive, or financial filing into chunks the way most LLMs require, SubQ is designed to process the whole document at once, using a proprietary sparse-attention method the company says cuts compute by two orders of magnitude at the 2-million-token mark compared with frontier models. The product is in early access: interested users apply for API access through a signup form, and no pricing is published yet. Benchmarks cited are the company's own, covering graduate-level science, competitive programming, and financial-automation tasks.
- Platforms
- API
- API
- Yes
- Last verified
- Oct 15, 2026
Is Subquadratic legit?
We don’t label tools safe or unsafe. Below are the records we checked ourselves, each with the date we read it.
- Domain registered
- Jan 7, 2026Registry record (RDAP), registrar NameCheap, Inc., looked up Sep 15, 2026
- TLS certificate
- Issued by Amazon, valid until Apr 5, 2027Seen in our own connection to the site on Oct 15, 2026
- Last checked
- Oct 15, 2026The site answered
- On the market since
- 2026The vendor's own material, read by us
- Operated by
- Subquadratic Inc., United StatesOur research, card verified Sep 1, 2026
Pros & cons of Subquadratic
Subquadratic is betting that a purpose-built sparse-attention architecture beats chunking-and-retrieval for genuinely huge documents, backed by real seed funding and a team with frontier-lab pedigree. Every performance and efficiency number available right now comes from the company itself, and the product isn't broadly available yet, so it can't be evaluated hands-on.
Pros
- Raised $29M in seed funding, per independent press coverage, backing continued development
- Claims a large compute-efficiency advantage at multi-million-token context via a proprietary attention method
- Founding team has direct frontier-lab and relevant enterprise AI experience
- Targets a genuine gap: true whole-document reasoning versus chunk-and-retrieve workarounds
Cons
- Not generally available; access is gated behind an application
- No published pricing
- All benchmark numbers currently come from the company itself, not independent evaluation
Best for
- Teams that specifically need whole-document reasoning over multi-million-token inputs (huge codebases, contract archives)
- Organizations willing to work with an early-access API rather than a GA product
- Technical evaluators comfortable validating vendor-supplied benchmarks themselves
Not for
- Anyone needing a generally available, self-serve API today
- Use cases well served by standard-context models with retrieval
- Buyers who need published pricing before evaluating
Watch out: Efficiency and accuracy claims (such as the cited compute-reduction figure) are self-reported pre-GA numbers; confirm against your own workload before relying on them.
Alternatives to Subquadratic
Subquadratic head-to-head
Side-by-side fact sheets against the competitors our research or their makers name — pricing, platforms, API and both verdicts on one page.