AI tools ·

SpaceXAI launches Grok 4.7, a longer-horizon coding and knowledge-work model

SpaceXAI released Grok 4.7 on September 21, a new frontier model built around longer task runs: a larger base model and a longer reinforcement-learning cycle on multi-hour task mixes, tuned to stay on hard coding and knowledge-work assignments and check its own output over long runs. It carries a 500k-token context window, takes text and image input, and offers four reasoning-effort levels — low, medium, high (default), and xhigh — so users can trade speed for depth.

The notable move is economic rather than architectural: pricing is held at Grok 4.6 levels, $2 per million input tokens and $6 per million output tokens below 200k prompt tokens ($4 input / $12 output above that threshold, with cached input discounted to $0.50). A Grok 4.7 Fast variant runs at twice the output speed for twice the price, but it is available only in Cursor and Grok Build — not on the public xAI API. Access is now open in Cursor, Grok Build (which offers free trial access), the Grok API, third-party coding harnesses, model routers, and cloud platforms.

SpaceXAI-reported benchmarks include 46.3% on CursorBench 4.0, 71.0% on DeepSWE v1.1, 64.0% on EEBench, 38.0% on Terminal-Bench 4.0, 19.6% on the Harvey Legal Agent Benchmark, and a score of 1,657 on AA Briefcase v1.1 — all first-party figures, not independently rerun. The company also says select cybersecurity partners are getting invite-only access to its red-team capabilities for defense research.

Why it matters

The competitive axis here is completed-task economics at a fixed price: SpaceXAI shipped a bigger, longer-horizon model without raising token rates, and Grok Build's free trial makes it the cheapest way this week to hand a long coding task to a frontier-class agent. For builders choosing between premium-priced APIs and this, the per-task cost math just shifted.

Key facts

  • Released September 21, 2026; pricing held at Grok 4.6 levels — $2 / $6 per million input/output tokens below 200k context
  • 500k-token context, text + image input, adjustable reasoning effort (low / medium / high / xhigh)
  • Live in Cursor, Grok Build (free trial), the Grok API, third-party coding harnesses, routers, and cloud platforms
  • Grok 4.7 Fast: twice the output speed at twice the price, only via Cursor and Grok Build
  • SpaceXAI-reported scores: 46.3% CursorBench 4.0, 71.0% DeepSWE v1.1, 64.0% EEBench, 38.0% Terminal-Bench 4.0
  • Invite-only red-team access for select cybersecurity partners for defense research

Sources

Spot an error? We correct quickly and note it. Contact us via the contact page.

← More AI news