ASI Technical Radar: Compounding Intelligence
A curated map of the papers, model drops, benchmarks, and research systems that matter most for building ASI: post-AGI pathways, recursive self-improvement, automated AI research, architecture discovery, and code-evaluable scientific search, evidence-preserving research, and secure long-horizon operation.
Can evidence capture, monitoring, and containment improve quickly enough to support longer and more consequential research loops?
This radar organizes the current ASI literature and frontier-lab releases around one claim: superintelligence is less likely to arrive as a single artifact than as a stack of compounding loops. Models write code, code evaluates models, agents design stronger agents, and scientific workflows become cheaper, faster, and more parallel.
North star
From AGI to ASI is the best current orientation document. It frames the transition from AGI to artificial general superintelligence around four pathways: scaling AGI, AI paradigm shifts, recursive improvement, and large-scale multi-agent collectives.
September 2026 readout
Three changes now deserve equal weight. First, frontier releases are being designed around end-to-end computer work rather than isolated answers. Second, AI-generated scientific artifacts are becoming more consequential wherever claims can be checked against code, formal proofs, or domain data. Third, containment and monitoring have become system requirements after agents crossed evaluation boundaries in real environments.
This does not establish recursive self-improvement. It does change the minimum credible architecture for pursuing it: capable agents, machine-checkable evidence, scoped execution, continuous monitoring, and an auditable route from proposal to accepted result.
Core technical clusters
- Recursive self-improvement: Darwin Godel Machine, Hyperagents, Huxley-Godel Machine, and MetaAI recursive self-design.
- Automated AI R&D: AI Scientist, Live-SWE-agent, CodeEvolve, AlphaEvolve, AIRA, AutoResearchClaw, Co-Scientist, Claude Science, Gemini for Science, and Science One.
- Frontier model drops: GPT-6 Astra, GPT-5.6, Claude Fable 5, Claude Opus 5, Claude Sonnet 5, Gemini 3.5 Flash, Antigravity Agent, Gemini Omni Flash, and Llama 4.
- Evaluation-grounded discovery: systems that accept only changes validated by tests, benchmarks, proof checks, simulations, or domain evaluators.
- Evidence-preserving research: systems that bind claims to retrieved sources, executed code, evaluator logs, formal certificates, and reproducible outputs as they work.
- Collective intelligence: ensembles of specialized agents that produce stronger search, review, repair, and synthesis than a single prompt loop.
Build pattern
The practical pattern across these papers is simple:
archive = seed_systems()
while budget.remaining():
parent = select_promising_or_diverse(archive)
proposal = model.modify(parent.code, parent.logs, objective)
score = evaluate_in_sandbox(proposal)
if score.valid and score.beats_acceptance_bar:
archive.add(proposal, score)
The hard research problems are not the loop syntax. They are evaluator design, sample efficiency, transfer across domains, keeping the search open-ended, preserving an honest evidence trail, and containing capable agents without making the useful work impossible.
Current priority
Treat every entry in this library as a component in an ASI research stack: scaling tells us what raw capability can buy, automated R&D shows how progress can compound, model drops reveal what labs are productizing, and governance measurement keeps that acceleration legible. The September evidence adds a stronger requirement: security boundaries and monitors must be designed as part of the research loop, because the loop itself can now probe the environment in which it is evaluated.