BREAKING
SpaceXAI Releases Grok 4.6
0Elo
GDPVal-AA v2
0Elo
AA-Briefcase
0%
Harvey LAB
Legal Work Remains Difficult
Grok 4.615.8
Grok 4.512.9
Fable 511.3
GPT-5.6 Sol2.5
A Longer Agent Training Run
1Supplemental training
2Regenerated SFT
3Agentic RL
4Self-testing
Competitive, Not Universal
StrengthsLaunch gains
Long-horizon knowledge work
Coding and visual projects
CaveatsReview required
Uneven benchmark leadership
Low absolute legal completion
Production Evidence Comes Next
AI NEWS BLITZ
SpaceXAI is shipping Grok 4.6 with a focus on sustained agent work.