Three model cards, three evaluation questions
An initial evaluation shortlist across text, speech, and multimodal tasks. These are selected models, not a ranking or a list of new September releases.
Super Intelligence Newsletter
Where AI capability is heading, read at the source, with the question worth testing on your own work.
Super Intelligence Newsletter
The frontier of AI, once a month.
The models, research, and releases that change what a business can build, with the sources and the question worth testing.
The newsletter
One email a month on the frontier of AI.
The Super Intelligence Newsletter follows how AI capability is moving: new models, research, and releases. Each issue picks the developments that change what a business can build.
It is written for people who decide what runs inside their business. The aim is a clear next step, not a news digest.
An issue does not ask you to change anything. It helps you decide what deserves a test on your own work.
Every issue is reviewed by an engineer before it is published. Draft issues stay unlisted until that review is complete.
Issues
Inside each issue
Four parts, in the same order every month.
What changed
The capability shifts of the month that matter for business work, stated as the publishers state them.
Candidates to test
A short list of models or tools worth evaluating, with the task each one suits and what to measure.
Sources and terms
A link to each primary source, and a note on licenses and terms. Record the exact version and terms before any deployment.
The question to test
For each candidate, the question your workload should answer, such as quality on your documents or speed on your hardware.
Sourcing rules
Primary sources only.
Each issue describes what publishers state, and links to where they state it.
- Facts come from the publisher’s own model card, paper, release notes, or license page.
- Each item links to its official sources, so you can check every statement.
- Any figures a publisher reports are labeled as reported by the publisher.
- Issues contain no Sophrono benchmark results.
- Issues contain no rankings, scores, or forecasts.
Using an issue
From a development to a decision you approve.
- Pick a bounded taskChoose one piece of work with clear inputs and a result you can check.
- Agree the testWrite down the examples, the threshold, and who will judge the result before anything runs.
- Compare under the same conditionsRun each candidate on the same examples, hardware, and settings, and record what you measure.
- A person approves any changeA named person on your side reviews the evidence and approves, or declines, the change.
The current approach stays a valid answer. A new capability earns a change only when it meets the agreed test.
Who it is for
Owners and technical leads deciding what to run.
- Owners and managers deciding where AI fits in their business.
- Technical leads who need primary sources and license terms before a test.
- Teams with sensitive records weighing vendor models against models they run themselves.
- Anyone who wants one considered read a month instead of daily release news.
Where it leads
When a development applies to your work.
Have a workload in mind? Evaluate it for AI, free.
Questions
What is the Super Intelligence Newsletter?
A monthly email from Sophrono engineers on how AI capability is moving. Each issue covers the developments that change what a business can build, links the primary sources, and names the question worth testing.
Does it predict when superintelligence arrives?
No. It tracks what changed, what the publishers state, and what that means for work you run today. It contains no forecasts or hype.
How often is it published?
Once a month, by email and on this page. An issue is published after an engineer has reviewed its sources.
Does it cover open and closed models?
Yes. It covers models you call through a vendor and models you can run in your own environment. License and hardware notes appear where they matter.
Does it include benchmark figures?
Issues contain no Sophrono benchmark results. Where an issue mentions a figure a publisher reports, it is labeled as publisher-reported and linked to its source.
Is it a ranking?
No. A shortlist in an issue names candidates worth evaluating on your own work. It has no scores and no order of merit.
Can I suggest a topic or a model?
Yes. Ask an engineer or use the contact page. Suggestions are reviewed against the same sourcing rules, and inclusion is not promised.
Does an issue tell me what to change?
No. An issue frames the question to test. The answer comes from an evaluation on your own work, and a person on your side approves any change.
Test a model on your work