Kimi K3Moonshot AI · Jul 16, 2026
1Mtokens
An instrument panel for the accelerating future
Observe the forcesreshaping intelligence.
The evidence behind History's Future—made visible, comparable, and alive. Track the systems, test the scales, then follow the argument back to the book.
38.9072° N / 77.0369° WField record 2026.07 Observatory online
01
Field briefing
Kimi K3 is the newest tracked frontier release.
Moonshot AI leads the release tape as million-token context and long-horizon agents move from exceptional claims toward the expected frontier.
↓Models tracked10Frontier systems
Million-token models06Native or extended
Open-linked05Weights or repository
Primary sources10Source-backed records
New
Forecast ledger / living record
What we expect.
Where experts disagree.
Six falsifiable forecasts place probability, deadlines, contrary evidence, and revision history on the same page.
New
Work Horizon Lab / applied frontier
A job is a bundle
of unequal tasks.
Explore 60 official task statements across twelve occupations, then raise the reliability bar and add the friction of real workplaces.
02
Intelligence terminal
Read the field,
not the hype.
A source-backed monitor of frontier Chinese models and the capabilities that matter.
Compare with the U.S. frontier →Enter the autonomous frontier → Live dataset10 records / 10 sources
CompareModel / releaseContextSignal profileSource
GLM-5.2Z.ai / Zhipu AI · Jun 16, 2026
1Mtokens
MiniMax M3MiniMax · Jun 1, 2026
1Mtokens
ERNIE 5.1Baidu · May 9, 2026
—tokens
DeepSeek V4 ProDeepSeek · Apr 24, 2026
1Mtokens
DeepSeek V4 FlashDeepSeek · Apr 24, 2026
1Mtokens
Qwen 3.6 PlusAlibaba Cloud · Apr 9, 2026
200Ktokens
MiniMax M2.7MiniMax · Mar 18, 2026
200Ktokens
Qwen3-Coder 480B-A35BAlibaba Cloud · Jul 22, 2025
256Ktokens
DeepSeek V3.2DeepSeek · Dec 1, 2025
128Ktokens
03
Acceleration ledger
Claims that keep
moving after publication.
001
The Free Fall
AI inference costs fell 99.9% in under four years.
Intelligence is moving from a scarce service toward ambient infrastructure.
↗004The Clock Speed Problem
Machines iterate at silicon speed; institutions respond at biological speed.
The widening gap may matter more than any single model release.
↗008The Refinery Problem
India generates one-fifth of training data but owns a fraction of the compute.
The intelligence economy may repeat the resource asymmetries of the industrial age.
↗04
Scale laboratory
Biological time.
Silicon time.
Move the control to feel the clock-speed gap described in the book.
10×100,000×
One human workday8hours
becomes
Equivalent machine work1.14human-years
A scale model, not a forecast. It reveals why institutions built around meetings, terms, and annual budgets struggle to govern systems that can iterate thousands of times faster.
Read “The Clock Speed Problem”From the manuscript / History's Future
From the birth of the universe
to the birth of artificial intelligence.
The Observatory measures the change. History's Future tells the larger story—and these manuscript op-eds test its most provocative ideas in public.
Observatory field notes
Follow what changes
between editions.
Occasional updates on frontier models, autonomous systems, manuscript arguments, and new evidence added to the Observatory.