Counting down to January 2031, a published baseline moved only by the dated, sourced news below. Every adjustment is visible; nothing moves the clock silently.
Each card is one editorial judgment: the story, why it matters for the index, and exactly how many days it moved the date. Negative adjustments pull the index nearer; positive push it out.
GPT-6 Astra saturates FrontierMath Tier 4 at 97.6% and ExploitBench at 100%, and is the first model OpenAI rates Critical for cyber capability under its Preparedness Framework.
Finding unknown flaws and building working exploits without step-by-step human direction is a real autonomy jump. Held to -30 because the independent Artificial Analysis index first scored Astra level with its own predecessor.
From August 2, 2026 most remaining provisions of the EU AI Act apply, bringing high-risk system rules, governance requirements, and national regulatory sandboxes into force.
Compliance overhead slows frontier deployment in a major market; a modest outward nudge.
Moonshot AI published full weights for the 2.8-trillion-parameter Kimi K3, and DeepSeek V4 hit general availability with MIT-licensed variants scoring 80.6% on SWE-bench Verified.
Frontier-scale capability diffusing as open weights broadens the base of labs pushing the frontier.
Anthropic will deploy up to 2 GW of AMD Instinct MI450-series GPUs in Helios rack-scale systems starting in 2027, with AMD committing up to $5 billion in equity investment.
Multi-gigawatt capacity commitments keep training-compute growth on its steep trajectory.
GPT-5.6 Sol, Terra, and Luna went public on July 9 with state-of-the-art results across coding, knowledge work, cybersecurity, and science, after a Commerce Department review cleared broad access.
A new flagship posting state-of-the-art results across broad domains lands ahead of typical release-cadence expectations.
OpenAI limited GPT-5.6 to vetted preview partners at the US government's request, the first time Washington preemptively asked an American lab to hold back a model before release.
Pre-release government gating adds a new institutional checkpoint between frontier capability and broad deployment.
Anthropic's Mythos-class models launched June 9, citing months of engineering compressed into days at customers, state-of-the-art vision results, and novel drug-design work in the life sciences.
A new model class above the previous flagship tier, with enterprise results cited in production, is the strongest capability signal of the year so far.
FutureSearch's tracker finds the Metaculus community, Dario Amodei, and forecaster Peter Wildeford all moved their AGI dates later between 2025 and 2026.
When the aggregate of serious forecasters moves later, the index should track it; this is the anchor class the baseline is built on.
The baseline anchors to the median of public AGI forecasts when this index launched in August 2026: the Metaculus community question on the first general AI system, aggregate expert surveys, and published forecaster timelines. It is an editorial reference point, not a prediction; only the listed news items move the clock from it.
This is not a prediction market, a scientific forecast, or a promise. Nobody knows when, or whether, artificial general intelligence arrives, let alone the technological singularity some expect to follow it. The clock is a way of reading the news: a fixed anchor, explicit adjustments, and a source for every one of them, so you can disagree with any movement on the record.
Whatever the date turns out to be, AI can already do real work for you today. Describe the outcome and get a Blueprint: the right tools, the right prompt, and the steps to follow.
Build my route