2026-06
Orion-100B
A 100B-parameter model trained at $1.25 per hour.
Milestones on the line · WE ARE HERE
Cheap intelligence
The same job that cost a dollar a year ago now costs cents. That, not benchmark records, is why AI shows up in ordinary products.
The price of intelligence per unit of work falls faster than demand grows. This is the last milestone we can confidently call passed.
StatusWE ARE HERE
TypeMilestones on the line
Marker2026
What exists today8
Researched
2026-07-09
GPT-5.6 Terra and Luna
Terra for everyday work at half the cost, Luna for speed and price.
2026-07
Claude Sonnet 5
Long-run coding and tool use at $2 and $10 per million tokens through 31 Aug 2026.
2026-07-21
Gemini 3.6 Flash
The new generation of fast, high-volume models.
2026-07
Nemotron TwoTower
Open-weight diffusion LM: 2.42× throughput at 98.7% of baseline quality.
In progress
сейчас
Distillation and quantisation
Small models trained on big-model outputs take an ever larger share of everyday load.
Planned
в планах
On-device inference
A useful model in your phone with no internet — compute moves back from the cloud to the user.
Distant horizons
впереди
Intelligence as a utility
The point where the cost of a query becomes as unremarkable as the electricity for a light bulb.
Branches
Milestones on the line
To the line