Breaking

SpaceXAI Releases Grok 4.6 for Long-Running Agents at $2 per Million Input Tokens

August 12, 2026 at 12:34 EDT

Foundation Models
AI Agents
Software Dev & Coding

SpaceXAI released Grok 4.6 on Aug. 12, positioning its new flagship model for long-running agents, coding and knowledge work. The launch keeps API pricing at $2 per million input tokens and $6 per million output tokens while expanding the model's focus to sustained, multi-step projects.

Aug 12, 2026 · SpaceXAI

Grok 4.6 Goes Long

The flagship model targets sustained agent work, coding and polished interactive projects while keeping the standard API price unchanged from Grok 4.5.

AA Intelligence Index

Frontier, but not a sweep

Composite of nine benchmarks; higher is better.

Grok 4.5

Grok 4.6

GPT-5.6 Sol

Fable 5

Grok 4.6 gains five points over Grok 4.5, ties GPT-5.6 Sol and finishes one point behind Fable 5.

500K

token context window

per 1M input tokens

per 1M output tokens

130 TB/s

NVLink rack bandwidth

Training path

Built to stay with complex work

Supplemental training

→

SFT trajectories

→

Trace filtering

→

Agentic RL

What looks promising

Substantial first passes

Stronger app structure, visual work and self-checking over longer trajectories.

What remains open

Workflow proof

Early reactions are mixed, and benchmark gains still need direct, long-running production tests.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.