SpaceXAI unveils Grok 4.7, positioning it as a budget-friendly powerhouse for coding tasks

SpaceXAI has officially released Grok 4.7, its latest AI model specifically engineered for coding and professional knowledge work. The company positions the model as offering competitive performance at significantly reduced costs compared to rivals, pricing input tokens at just $2 per million and output tokens at $6 per million, substantially below competitors like GPT-5.6 at $4/$20 and Fable 5.1 at $10/$50. The release comes amid intensifying competition in the enterprise AI market, where price-performance ratios increasingly determine adoption.

According to the announcement, Grok 4.7 represents a substantial leap in technical capability, particularly for extended software engineering tasks. On CursorBench 4.0, a benchmark stressing longer-running coding operations, the model achieved 46.3% performance, outperforming GPT-5.6 Sol max (41.7%) and edging ahead of Grok 4.6 (40.4%). The model was trained on a more challenging task mix with longer reinforcement learning runs, enabling it to work through multi-hour office workflows and terminal sessions more effectively. Notably, while Fable 5.1 max leads on CursorBench at 51.8%, Grok 4.7 delivers its results at roughly one-fifth the cost per token.

Beyond raw performance, the company emphasized enhanced safety systems as a differentiator. The model incorporates a completely new safeguard stack that reportedly achieves the highest HackerBench v0.3 scores for balancing legitimate cybersecurity utility with refusal of dangerous requests, allowing only 3.3% of risky dual-use prompts through while rarely blocking benign security work. A fast variant doubles output speed at double the price. Grok 4.7 is available immediately through Cursor, Grok Build, the Grok API, and select cloud platforms, with invite-only red-team access offered to select cybersecurity partners for defense research.