SubQ 1.1
#25 v Small & Edge modelySubquadratic · v1.1 · 2× · naposledy 28. 9. 2026
24
Momentum
SubQ 1.1 Small is a language model from Subquadratic based on the proprietary "Subquadratic Sparse Attention" (SSA) architecture, which enables linear rather than quadratic scaling of computational effort with context length. It achieves near-perfect Retrieval results with context windows of up to 12 million Token and, according to the manufacturer, requires significantly less computing power than dense attention mechanisms. The model is currently being deployed with selected design partners; weights are not openly available. It focuses on text and code processing; multimodal capabilities (images, etc.) have not yet been released.
Vývoj momenta
30.06.28.09.
Vlastnosti
| Key Benchmark (%) | RULER 128K: 99.12%; Needle-in-a-Haystack 1–2M: 100%; 6–12M: 98% |
| Context Window (Tokens) | up to 12,000,000 tokens (near-perfect retrieval at 1M, 2M, 6M and 12M) |
| License | Closed-weights, not open source; based on an open-weight donor model modified with SSA |
| Multimodality | Text/code-only; no published multimodal benchmarks (no documented image input) |
| Platform | API (OpenAI-compatible), currently private beta with design partners |
| Release Date | June 16, 2026 (model card / technical report) |