Overview

Grok is xai/spacexai’s family of frontier language models integrated with the X platform, Grok Build, cursor, grok-bot, and government deployments via grok-gov.

Agent governance incident (October 2026)

Government & Policy Use (October 2026)

Venezuela details rely on single anonymous source. Outcomes (ongoing custody, no free elections) contradict optimistic prediction.

Grok 4.7 (September 21, 2026)

  • Availability: cursor, Grok Build, SpaceXAI API, third-party coding harnesses (2026-09-21-grok-4-7-release)
  • Pricing: 6/M output (unchanged from 4.6); fast variant 12
  • Capabilities: Long-running coding/knowledge work; 256k standard / 500k long context; xhigh/high/medium/low effort levels
  • Benchmarks: Claims frontier price-performance on cursorbench 4.0, DeepSWE v1.1 — vendor-reported

Benchmark scores (CursorBench 4.0, DeepSWE) are vendor-reported. Independent replication pending.

Grok Bot (August 11, 2026)

  • Persistent always-on agents on shared cloud VMs (grok-bot)
  • Grok 4.6 tuned for long-running agentic tasks
  • Bundled with SuperGrok Heavy, Cursor Ultra (120/seat)

Grok 4.5 Public Launch (July 8, 2026)

  • Availability: Grok Build, all Cursor plans, SpaceXAI API console (2026-07-09-grok-45-xai-official)
  • Pricing: 6/M output; fast variant 18
  • Architecture: MoE; co-trained with cursor on NVIDIA GB300 cluster
  • EU: Not available at launch; expected mid-July 2026
  • Claims: Opus-class performance, ~4.2× fewer output tokens on SWE Bench Pro vs Opus 4.8 max — vendor benchmarks

Benchmark claims from SpaceXAI/Cursor announcements; independent verification limited.

Grok 4.5 Private Beta (June 2026)

Release Cadence

spacexai plans continued model releases; Grok 4.5 is first public post-IPO model.

Sources