LLM

Continuous context. Linear compute.

Continuum1-9B is Innomium foundation model for extreme long-context reasoning.

The value stack

Hybrid GLA + Gated DeltaNet with NOPE.

Structural distillation + 2M-context expansion.

continuum-flash-linear-attention Triton kernels.

Reasoning, math, and long-document QA at 2M context.

Linear compute at million-token scale.

75% MMLU, 37% MATH-500.

Continuum1-9B on Hugging Face.

Innomium flash-linear-attention fork.

llm

Fully linear foundation model for extreme long-context reasoning.

~8.6B2M tokens75.0%

Completed Jun 2026 · 41 participants · 3 PRs merged

$8,000 prize

llm

A fully linear 8.6B foundation model with 2M native context — hybrid GLA + Gated DeltaNet, open weights, and production kernels on Hugging Face.

Innomium LLM Team · 3/1/2026

llm

Quadratic attention does not scale to million-token workloads. Continuum's linear layers keep latency predictable for production pipelines.

Innomium LLM Team · 2/20/2026

~8.6B params, 2M context, BF16 on Hugging Face.