§ 1 · Profile
What it is good at
Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.
Agents52.2−15.9—/267✱0/5
Tool use55.6−18.4—/81✱0/1
Reasoning47−19.9—/178✱0/4
Mathematics47.2−18.5—/140✱0/2
Science38.9−24.6—/122✱0/1
Expert reasoning41−19.9—/0✱✱
Coding45.6−23.5—/164✱0/5
Agentic coding44.5−24.2—/157✱0/4
Code generation38.1−31.9—/77✱0/2
§ 2 · Head to head
What it beats, and what beats it
The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for G9v3 3B, left for the other.
§ 3 · Sources
Where the numbers come from
1 publication, 7 figures. Every one links to the page it was read from.
Terminal-Bench 2.1 6%SciCode 17.7%GPQA Diamond 43.8%SWE-bench Verified 16.4%HMMT Feb 2026 34.1%HLE 4.5%BFCL v4 47.9%
Badge
[](https://publicai.io/model-index/m/g9v3-3b)