Post
The Top 5 SI Frameworks Every Developer Must Know
Five framework choices for developers building on language models, and why “must know” is not a performance ranking.
Disclosure. Further reading links to a sister publication under the same ownership. Those pages are commentary, not evidence for a score or a ranking on this site.
The short answer
Developers should be able to name the model API, the orchestration layer, the evaluation harness, the permission boundary, and the log. Those are categories, not a top-five product ranking. We have not benchmarked agent frameworks.
How to read this list
This is an editorial map, not a tested ranking and not a score. BestSuperintelligent has not independently tested the products, discoveries, or headlines named here. Nothing below is a winner. Use it to decide what to verify, then check a primary source.
Framework lists go stale as soon as a vendor renames an SDK. Learn the seams, then pin versions.
Which five seams matter more than a brand list?
- The model call itself: price, context, and the version string.
- Orchestration: who decides the next tool call.
- Evaluation: the tests you run, not the vendor’s demo.
- Permissions: files, network, and secrets the agent can reach.
- Logs and replay: enough to explain a bad change.
Our planned comparisons, including GitHub Copilot vs Cursor, stay unscored until a protocol is run. Choosing a coding assistant tells you how to trial one on your code.
Evidence in this article Editorial interpretation Our reading of what the evidence means. Judgment, clearly separated from the evidence itself.
Frequently asked questions
What is the best SI framework?
We are not naming one. Pick from the job: a hosted coding agent, a library you run, or a workflow engine, then test it on your repo.
Where do Copilot and Cursor fit?
They are products in our planned coding comparison, not finished rankings. The questions are on the compare page. No winner is published.
Should I learn one framework brand?
Learn the five seams first. Brand SDKs change name faster than the job of logging and permissions.
Sources
- How to read an evaluation BestSuperintelligent. Our own standard for what a result must show.
- Methodology BestSuperintelligent. Proposed weights and the rule that missing evidence is not a score.
Further reading
From a related publication under the same ownership. It is commentary, not evidence for anything on this site.
- Ten open-weight models to research, TheBestSI (sister site). Sister-site guide. Weights and licences should be checked on the publisher’s page.