BREAKING
SpaceXAI Ships Grok 4.6
0
K tokens
context window
0
$/M
input
0
$/M
output
DeepSWE Shows a Clear Gain
Grok 4.5
54
Grok 4.6
65.9
GPT-5.6 Sol Max
73
Built for Longer Agent Runs
1
Research
↓
2
Build
↓
3
Verify
↓
4
Refine
Capability Meets Caveats
Stronger focus
Agent work
●
Longer task trajectories
●
More self-testing
Open questions
Early release
●
Company-reported scores
●
Higher long-prompt rates
Independent Tests Come Next
AI NEWS BLITZ
SpaceXAI has released a new model focused on sustained agent work.