Choose a task
Release News
CodingCircuit board designWeb researchScientific workComputer use
Explore
Results matrixPrice vs abilityRelease watch
Evidence base
20 Nov 2026

25 frontier labs tracked · first-party sources only.

Tasks/Coding

Coding matrix

Compare the work models can do, then see what changed.

Overall average

A winner needs the complete picture.

The layout is ready. Complete the benchmark set and verify comparable runs before awarding the crown.

Click any score to view source proofPrice vs ability
Model
Model provider
Terminal-Bench 4.1
Tasks resolved
FrontierCode 2.0
Weighted rubric score
FrontierSWE v2
Issues resolved
Average
Equal weight

No verified first-party benchmark results published in coding yet.

As Eve detects official model cards and technical reports (or when you run a lab backfill in admin), verified scores appear here with direct source links.

Complete the comparison

Every benchmark in this basket carries equal weight on a 0–100 scale. A model must report all 3 benchmarks under comparable settings to earn an overall rank.

First-party announcements, technical reports, and model cards only. Different effort and agent settings remain labeled; scale alignment does not equalize benchmark difficulty.

Latest in coding

Provider updates

Release news, relevant results, and a link to the first-party source.

xAI·Sep 21, 2026

Introducing Grok 4.7

CursorBench 4.0: 46.3% at xHigh; DeepSWE v1.1: 71.0% at high effort; Terminal-Bench 4.0: 37.6% at xHigh (not Terminal-Bench 4.1).

Provider report

Release watch

Days since launch, against this family's historical interval.

Grok 4.7

Released Sep 21, 2026
60
days since release
—
days average interval
First Grok release · no interval in this series
4.7 / SEP 21· History starts here

Average = mean days between consecutive public releases in the same family. Historical cadence is not a release forecast.

Best Model/First-Party Evidence Only/20 Nov 2026
Release newsAll release cadences