UofAi

AI Pulse

The honest read on AI progress

Three transparently-sourced dials: how much raw compute is behind the frontier, how far today's systems are from an operational definition of Super AI, and which human milestones AI has reached. Every number shows its source, retrieval date, and data class — expand any dial to check our work.

UCP
239.8

UofAi Compute Pulse

as of 2026-07-02

Components & sources (4)
  • Largest known training run500,000,000,000,000,000,000,000,000FLOP
    Epoch AI — Trends dashboardretrieved 2026-07-02Estimated

    Largest publicly estimated run (Grok 4). Newer frontier runs likely exceed this but are undisclosed. Frontier training compute grows 4–5× per year.

  • Global AI chip compute stock growth3.4× per year
    Epoch AI — Trends dashboardretrieved 2026-07-02Estimated

    Doubling roughly every 6.8 months. Largest single AI data center ≈ 800,000 H100-equivalents.

  • AI chip performance per dollar37% per year improvement
    Epoch AI — Trends dashboardretrieved 2026-07-02Measured

    Inference cost at fixed capability has been halving roughly every 2 months.

  • Frontier training power draw100MW+ per frontier run
    Epoch AI / EPRI — Scaling Intelligenceretrieved 2026-07-02Extrapolated

    Growing 2.2–2.9× per year; single runs projected to reach 4–16 GW by 2030.

SHI
0100
55.7

band 51.060.0

Superintelligence Horizon Index

as of 2026-07-02

Components & sources (3)
  • Expert knowledge frontier (Humanity's Last Exam)59.2

    VERIFY: reported SOTA ranges 45.7–53.3% across leaderboards as of 2026-06-30; Scale/CAIS official leaderboard is canonical. Human domain experts ≈ 90%. Range published per UBS-1 conflict rule.

  • Novel-skill acquisition (ARC-AGI-2 & 3)33.3
    ARC Prize Foundation (semi-private eval)retrieved 2026-07-02Measured

    ARC-AGI-2 SOTA 54% (verified Dec 2025); ARC-AGI-3 best purpose-built agent 12.58% at March 2026 launch, frontier LLMs <1%. Human baseline 100% on both.

  • Autonomous task horizon (METR 50%-reliability)74.5
    METR — Task-Completion Time Horizonsretrieved 2026-07-02Measured

    Frontier 50%-horizon ≈ 12–16 hours (May 2026); METR notes measurements above 16 hours are unreliable with the current task suite. Target for SHI=100 on this subscale: one work-month (167 hours). Doubling time ≈ 4 months in 2024–2026, up from 7 months historically.

HPT
0100
54.2%

Human Parity Tracker

as of 2026-07-02

Human Parity ledger

Landmark human capabilities AI has reached, is approaching, or has yet to reach — each with falsifiable criteria and a citation.

Coding & Software

  • Elite competitive-programming parity
    2025 · OpenAI reasoning modelsAchieved

    Gold-medal / top-rank performance at IOI/ICPC-level contests.

    Wikipediaretrieved 2026-07-02Measured
  • Autonomous resolution of most real-world software issues
    Frontier coding agentsFrontier

    Majority resolution on verified real-repository issue suites with human-level acceptance; strong but not yet reliable across arbitrary codebases.

    SWE-bench (public leaderboards)retrieved 2026-07-02Measured

Economic & Professional

  • Majority of remote-work tasks at expert quality
    Open

    Independent evaluation shows AI completing >50% of a representative remote-task basket at expert quality and cost parity.

    Curatedretrieved 2026-07-02Estimated
  • Award-competitive long-form creative work, judged blind
    Frontier

    AI-authored long-form work places in blind professional judging; strong short-form results exist, long-form remains contested.

    Curatedretrieved 2026-07-02Estimated

Games & Strategy

  • Superhuman chess
    1997 · IBM Deep BlueAchieved

    Defeats reigning human world champion in match play.

    Historical recordretrieved 2026-07-02Measured
  • Superhuman Go
    2016 · DeepMind AlphaGoAchieved

    Defeats top-ranked human professionals in even matches.

    Wikipediaretrieved 2026-07-02Measured
  • Superhuman multiplayer no-limit poker
    2019 · Pluribus (CMU/Meta)Achieved

    Beats elite professionals in six-player no-limit Hold'em.

    Wikipediaretrieved 2026-07-02Measured

Language & Knowledge

  • Broad academic exam parity (MMLU-class)
    2023 · GPT-4-class modelsAchieved

    Exceeds expert-human accuracy on broad multiple-choice academic suites; now saturated above 88–90%.

  • Passes controlled Turing-test studies
    2025 · GPT-4.5 (UC San Diego study)Achieved

    Judged human more often than actual humans in a controlled three-party Turing test.

    arXivretrieved 2026-07-02Measured
  • Expert parity on Humanity's Last Exam
    Open

    SOTA reaches ~90% (human domain-expert average); current SOTA 45.7–53.3% (VERIFY per §6.2 conflict note).

    Humanity's Last Exam (CAIS/Scale)retrieved 2026-07-02Measured

Long-Horizon Agency

  • One-workday autonomous task horizon
    2026 · Frontier agents (METR-measured)Achieved

    50%-reliability time horizon ≥ 8 hours on METR methodology; frontier ≈ 12–16 hours as of May 2026.

    METR — Task-Completion Time Horizonsretrieved 2026-07-02Measured
  • One-workweek autonomous task horizon
    Open

    50%-reliability horizon ≥ 40 hours; METR's current suite cannot yet measure reliably beyond 16 hours.

    METR — Task-Completion Time Horizonsretrieved 2026-07-02Measured
  • One-workmonth autonomous task horizon
    Open

    50%-reliability horizon ≥ 167 hours (the SHI autonomy target).

    METR — Task-Completion Time Horizonsretrieved 2026-07-02Extrapolated

Mathematics

  • International Math Olympiad gold-medal performance
    2025 · OpenAI & Google DeepMind systemsAchieved

    Official gold-medal-level score on IMO problems under competition conditions.

    Wikipediaretrieved 2026-07-02Measured
  • Research-level mathematics (FrontierMath majority)
    Open

    Solves a majority of FrontierMath research-tier problems.

    Epoch AI — FrontierMathretrieved 2026-07-02Measured

Novel-Skill Acquisition

  • Human parity on ARC-AGI-2
    Open

    ≥85% on the semi-private set at human-comparable efficiency; SOTA 54%.

    ARC Prize Foundationretrieved 2026-07-02Measured
  • Human parity on ARC-AGI-3 interactive reasoning
    Open

    Matches the 100% human baseline on unseen interactive environments; best system 12.58% at launch, frontier LLMs <1%.

    ARC Prize Foundationretrieved 2026-07-02Measured

Perception

  • Human parity in image classification
    2015 · ResNet-era modelsAchieved

    Surpasses human error rate on ImageNet.

    Wikipediaretrieved 2026-07-02Measured
  • Human parity in conversational speech recognition
    2017 · Microsoft ResearchAchieved

    Matches human transcription error rate on Switchboard.

    Wikipediaretrieved 2026-07-02Measured

Physical & Embodied

  • Broad driverless deployment
    Waymo and peersFrontier

    Fully driverless ride-hailing at scale in multiple metros; not yet generalized across all conditions/geographies.

    Curated (public deployment data)retrieved 2026-07-02Measured
  • General household manipulation
    Open

    A commercially available robot reliably performs varied household chores in unmodified homes.

    Curatedretrieved 2026-07-02Estimated
  • General-purpose humanoid in commercial work
    Open

    Humanoid robots performing economically useful multi-task labor at scale without teleoperation.

    Curatedretrieved 2026-07-02Estimated

Science & Research

  • PhD-expert parity on Google-proof science Q&A
    2024 · Frontier reasoning modelsAchieved

    Exceeds the 65% PhD-expert baseline on GPQA Diamond; SOTA now >90%, approaching saturation.

  • Autonomous novel scientific discovery
    Multiple lab systemsFrontier

    An AI system independently generates and validates a novel, peer-review-published scientific finding end-to-end; current systems co-author and accelerate but require human direction.

    Curated (multiple)retrieved 2026-07-02Estimated

Learn to work with the frontier

15 minutes a day, on real tasks.