ANALYSIS
Methodology note: Historical index values on this Pulse are normalized to the current methodology for comparability. The original as-published record remains unchanged in the canonical archive.
Verdict: no verified flagship/frontier model release in the strict 24-hour window changed the intelligence ceiling. The strongest movement is outside raw model capability: OpenAI published first-silicon Jalapeño inference results, Figure exposed the scale of its global physical-data pipeline, Accelerated Understanding disclosed a physics-native foundation-model program, and Pacific Fusion began construction of a net-facility-gain demonstration campus. These are meaningful enabler and execution signals, but none crosses a current maturity anchor.
Frontier intelligence / reasoning / agents / coding / AI-R&D
Two-pass release checks across OpenAI, Anthropic, Google DeepMind, xAI, DeepSeek, Qwen/Alibaba, Mistral, Meta and Cohere found no new flagship/frontier model launch in the strict window. Intelligence remains 63.7/100.
OpenAI did publish measured first-silicon results for Jalapeño, its custom inference accelerator. Across GPT-OSS 120B, DeepSeek R1 670B and Kimi K2.5 1T, OpenAI reports 1.5–1.9× more AI work per watt at peak throughput and 1.7–3.6× lower end-to-end latency than comparison systems on SemiAnalysis InferenceX operating points. OpenAI also reports that Codex with GPT-Astra brought three previously unsupported open-weight model families to high performance within two months, with selected AI-generated kernels 1.5–1.8× faster than prior human-expert implementations. This is a strong compute/AI-R&D loop signal, but the hardware measurements are still first-party and do not by themselves raise Intelligence maturity.
Compute & Infrastructure
OpenAI Jalapeño — relevance 8.6/10, evidence 91/100. The strongest new systems signal. Working silicon, matched-user-experience benchmark results and a deployment plan for OpenAI infrastructure by year-end materially strengthen the inference-efficiency trend. Compute remains 75.1/100 because the current Economics & Efficiency anchor already assumes rapid per-watt gains; production qualification and independent reproduction at scale remain the next evidence threshold.
Robotics / physical autonomy
Figure Index — relevance 8.1/10, evidence 86/100. Figure says its four-month stealth pipeline has collected 16M+ real-world task videos from 108 countries, with 44,000+ weekly active users, processing about 30 minutes of new video per second. The company plans more than $1B of data-and-compute spending over the next 12 months and says internal Helix generalization results are improving. This is unusually strong evidence of a scaling data engine for general-purpose robotics, but Figure has not yet published the corresponding capability benchmark, intervention-rate reduction or zero-shot task-performance delta. Robotics therefore remains 47.4/100.
AI for Science
Accelerated Understanding physical-AI foundation model — relevance 8.2/10, evidence 84/100. The company reports hundreds of pre-training runs up to 1T parameters, architecture scaling tests up to 35T parameters and a physics-native 4D model designed to replace slow simulation/experiment loops with learned physical prediction and optimization. It also claims context scaling beyond 5T physical tokens. This is a potentially important scientific-computing architecture, but independent benchmarking, reproducible task suites and prospective experimental validation are still missing. AI for Science remains 59.0/100.
Open Problem Closure Radar
The strict arXiv/VibeMathed/UnsolvedMath pass found no new qualifying AI-linked named-problem closure or independently verified status upgrade in the 24-hour window. Several older August conjecture-resolution papers surfaced in search, but their primary submission dates fall outside this Daily window and no new verified AI role was established for them. No Science maturity change is warranted.
Energy / Fusion
Pacific Fusion Albuquerque campus — relevance 7.8/10, evidence 90/100. Pacific Fusion broke ground on a roughly $1B research and manufacturing campus intended to host its first demonstration system targeting net facility gain, alongside an NNSA memorandum of understanding. This is genuine project-execution evidence, not a plasma-performance result. Energy/Fusion remains 46.8/100 until construction converts into integrated hardware, shot data and ultimately facility-level energy-gain evidence.
Synthetic Biology / Advanced Materials / Longevity / BCI-HMI
No new verified result in the strict window crosses a maturity anchor. Synthetic Biology remains 66.8, Advanced Materials 62.4, Longevity 36.3, and BCI/HMI 42.8.
Benchmark observatory
No new METR Time Horizon, Humanity's Last Exam or FrontierMath point requires ingestion or score adjustment. The most material benchmark-like evidence today is infrastructure-side InferenceX rather than a new frontier capability ceiling.
Readiness
ASI READINESS: 59.03/100 — Δ0.00
SINGULARITY READINESS: 53.19/100 — Δ0.00
ASI FORECAST: 5Y 30% | 10Y 81% — Δ0 pp
Central estimate: 2032–33 — unchanged.
Bottom line: the day strengthens a pattern already visible across the observatory: the frontier is broadening through faster inference, larger physical-world data engines, physics-native models and real infrastructure execution. Those developments can accelerate future capability, but today they do not yet demonstrate a higher intelligence ceiling, a closed autonomous science loop or substantially more general physical autonomy.