I would absolutely say you have non-zero fluid intelligence in your reasoning system.
“The number of teams in the world that I've seen that could produce a reasoning system with non-zero fluid intelligence ison the order of tens... most of all the frontier reasoning stuff is at frontier labs... I do not know of too many other people who built a general purpose reasoning system outside of a frontier lab that have done this.”
Mike Knoop
co-founder of Zapier · co-founder of ARC Prize and Ndea (with François Chollet) on Reasoner, evaluating an ARC-AGI-2 task on camera, May 2026
Each era's platform binds to its scarcest substrate.
AI's scarcest substrate is high-bandwidth memory (HBM). Context windows are bounded by it.
Era
Platform / OS
Scarcest substrate
Personal Computing
Microsoft Windows
Compute cycles
Mobile
Apple platform
Power-efficient compute (ARM · Apple Silicon)
AI
Reasoner · the AI context OS
High-Bandwidth Memory (HBM)
“The context data needed by large language models … can easily exceed their limitations, especially their context windows, which are imposed by memory walls, a physical constraint of datacenters. Hyperscaler efforts to drastically increase compute and memory capacities currently face diminishing returns, in no small part because of power-grid limitations.”
— Reasoner patent application 63/690,702 · written 2024 · Wayne Chang, sole inventor
Reasoner is the AI context OS — engineered for the HBM constraint, not in spite of it. Lossless-compressed knowledge cores that fit any LLM, any device, any environment.
Seven thesis pieces · five firms · five months. a16z said it most directly.
“Many names have emerged in discussion today — context OS, context engine, contextual data layer, ontology, and more — but the underlying concept remains the same.”
12 frontier LLMs average just 14.5%. The benchmark tests communication and social interactions, fragmented information & revisions, and behavioral records and activity trails — the messy, fragmented, multi-party context humans live in.
“Today's most advanced AI models are still struggling to understand our daily reality. Even when we give an AI our chat histories, scattered notes, or daily records, it still doesn't quite ‘get it.’”
— Tencent Hunyuan + Fudan University, “Real life is where context gets hard,” April 30 2026 (senior author Yao Shunyu, ex-OpenAI research scientist)
This is the failure mode Reasoner is engineered to fix — by construction.
Sources: S078 (Tencent Hunyuan blog 100039, “Real life is where context gets hard”) · S075-S077 (companion CL-bench paper, arxiv 2602.03587) · S060 (Reasoner patent claims · anti-RAG, deterministic, by construction) · CF-73 / CF-74
Frontier-tier outcomes. A fraction of the capital.
19 architectural-class peer new-AI-labs raised ~$15.4B over the past 18 months — ex-OpenAI / ex-DeepMind / ex-Meta / ex-Salesforce founders. Reasoner did it on $8M.
~$15.4B
19 verified peer labs · 18 mo
$8M
Reasoner cumulative pre-seed
~1,925×
cohort capital vs ours
Closest MindSim comp
Simile · $100M Series A (Feb 2026) · Index Ventures lead · Karpathy + Fei-Fei Li angels.
Closest Patented.ai comp
Patlytics · $40M Series B (Apr 2026) · Next47 lead · ~$61M total · 40%+ AmLaw 100.
An active-duty NSA-affiliated Air Force Intelligence Colonel walked us through procurement — after watching MindSim profile her own characteristics on camera.
“I'm going to endorse something that I will no longer have a job for.”
Col. Nicole “Nikki” Stanley
active-duty NSA-affiliated AF Intel officer · posted at the Defense Language Institute · 4 years at the Pentagon prior · 16-min on-camera session
5 IC use cases Stanley named — MindSim digital twins for each
3 IC procurement doors she offered to open for us
DLI Tech Director
tech procurement lead
AETC Intel Officer
Air Education + Training Cmd
Pentagon contacts
“really deep pockets”
SOC 2 + HIPAA + GDPR compliant · FedRAMP-ready · no memory-layer comp has this vertical
ARC Prize and Ndea (with François Chollet)
