Manifest AI

Overview

Manifest AI is a post-transformer architecture startup profiled alongside Subquadratic in MIT Technology Review’s Aug 10 2026 piece on production-scale alternatives to dense attention. The company’s headline contribution is power retention — a published mechanism pitched as a drop-in replacement for standard attention. Unlike Subquadratic’s vendor-claimed 1,000× compute reduction (still awaiting independent audit), Manifest AI’s power retention is a real, published mechanism, not a marketing artefact — but production adoption of it in mainstream open-weights releases remains nascent.

Timeline

2026-08-12-AI-Digest — MIT Technology Review‘s Aug 10 piece profiles Manifest AI alongside Subquadratic as two startups pushing post-transformer architectures at production scale. Manifest AI is shipping a “power retention” mechanism pitched as a drop-in replacement for attention; both companies argue dense attention has become the bottleneck as context and model size grow, and both frame recent reasoning-model advances as workarounds patching over transformer flaws. Narrow read the corpus carries: Manifest AI’s power retention is a real, published mechanism, not a marketing artefact — but production adoption is still nascent. Structural read: the useful framing is not “post-transformer moves from curiosity to product” (Mamba / RWKV had that framing in 2023) but “another funding data point in the multi-year drift toward hybrid subquadratic stacks” — the June 2026 “On Subquadratic Architectures” survey still frames the field as principle-seeking, with hybrids (Samba, Nemotron Nano, Kimi Linear, Olmo Hybrid) replacing some attention layers rather than full replacement. 30 / 60 / 90-day watch: whether power retention gets integrated into a mainstream open-weights release; whether long-context evals (>1M tokens) surface measurable hybrid-vs-attention deltas.

Key Developments

Power Retention as Drop-In Attention Replacement (August 10, 2026): Manifest AI’s published mechanism is framed as a drop-in replacement for standard attention — the specific technical claim is real and published (unlike Subquadratic’s vendor-only 1,000× headline), but production adoption in mainstream open-weights releases remains nascent. The corpus test is whether power retention appears in a second, independent frontier-scale open-weights release inside 90 days.
Named Alongside Subquadratic in MIT TR’s Post-Transformer Product-Scale Framing: The Aug 10 MIT Technology Review piece pairs Manifest AI with Subquadratic as the two named startups pushing post-transformer architectures at production scale — two more funded companies is signal but not a shift, since deployed inference workloads are still attention-dominated and hybrids (partial attention replacement) rather than full-replacement architectures continue to be the practical default.

2026-09-01-AI-Digest — MIT Technology Review profiles Manifest AI in a piece on the wave of startups exploring alternatives to attention to break the quadratic-context cost curve (MIT Technology Review) — Manifest AI’s “power retention” architecture (rolling-summary variant, closer to the retention/state-space family than a strict hybrid), shipping PowerCoder (StarCoder variant) and Brumby (with claimed parity to open Qwen variants). Narrow read the digest carries: the framing of “approaching production viability” is behind the field — AI21 Jamba 1.6, Mistral Codestral Mamba, IBM Granite 4.0, and Google RecurrentGemma have all shipped SSM / hybrid production models, and hybrids at ~43% SSM + ~7% attention are matching or beating pure Transformers on benchmarks. Structural read: attention alternatives are already shipping in production, not just approaching it — Manifest AI is the research-demo end of an already-live category. Watch clause: does Manifest AI ship a model that clears the open-Qwen bar on independent evals, or does the parity claim stay internal? Log against MOC - Open Source Models.

Manifest AI

Overview

Timeline

Key Developments

Related