BREAKING
SpaceXAI Releases Grok 4.6
0
K tokens
Context
0
$/1M in
Input
0
$/1M out
Output
Grok 4.6 Reaches 61
Grok 4.5
56
Grok 4.6
61
GPT-5.6 Sol
61
Fable 5
62
Training for Longer Runs
1
Supplemental training
↓
2
SFT regeneration
↓
3
Trace filtering
↓
4
Agentic RL
Coding Promise, Proof Pending
Promise
Strengths
●
Strong first-pass apps
●
Long-run self-checking
Caveat
Evidence
●
Mixed hands-on reports
●
Benchmarks need validation
Real Workloads Decide
AI NEWS BLITZ
SpaceXAI has released a new flagship focused on long-running agent work.