SpaceXAI released Grok 4.7, a new flagship model the company positions for coding and knowledge work, priced and served at the same rate as its predecessor. The model "works longer on difficult tasks, checks its own work more carefully, and comes with our best-calibrated safeguards to date," according to SpaceXAI's announcement. Pricing stays unchanged, SpaceXAI says: $2 per million input tokens and $6 per million output tokens, with a faster variant available at twice the output speed for twice the price. On its own benchmark table, SpaceXAI reports the model at 46.3% on CursorBench 4.0, ahead of Grok 4.6's 40.4% and GPT-5.6 Sol's 41.7%, though behind Fable 5.1 at 51.8%; on DeepSWE v1.1 it scores 71.0% (marked high-effort) against Grok 4.6's 65.2%; on EEBench it scores 64.0% against Grok 4.6's 53.0%; and on Terminal-Bench 4.0 it scores 38.0%, well behind Fable 5.1's 57.9%. SpaceXAI also reports a GDPval Elo score of 1,695 for the model, behind the same rival's 1,735 and ahead of the earlier model's 1,605 and GPT-6 Astra's 1,542. All of these are SpaceXAI's own reported figures, without independent verification.
SpaceXAI says the model runs on a larger base model than its predecessor, trained with a longer reinforcement-learning run weighted toward tasks that take many hours to complete, and says it was trained to natively understand the Grok Bot harness for better conversational and general knowledge-work performance.
SpaceXAI says the model ships with an entirely new safeguard stack and describes it as the strongest it has tested on refusal and jailbreak resistance. The company says it tops a leading biosafety benchmark and allows only a small share of risky dual-use prompts through on a cybersecurity benchmark, while limiting false refusals on legitimate security work. SpaceXAI says it has also begun giving select cybersecurity partners invite-only access to the model's red-team capabilities for defense research.
The model is available now in Cursor and Grok Build, and through the Grok API, third-party coding harnesses, and model routers and cloud platforms, per SpaceXAI. The announcement does not state a rollout date for the faster variant or give an independent benchmark source to check the comparisons against.
