SpaceXAI Releases Grok 4.7: Larger Model at Same Price
SpaceXAI launches Grok 4.7, a larger base model priced at $2/$6 with improved Terminal-Bench scores and advanced safety features.
Stock photo for illustration only, not from the actual event
- SpaceXAI introduces Grok 4.7, a larger base model keeping the exact pricing of $2 per million input and $6 per million output tokens.
- Terminal-Bench 4.0 score jumps significantly from 20.3% to 38.0%, outperforming its predecessor across all evaluated metrics.
- Available immediately as a hosted model via xAI API, Cursor, Grok Build, OpenRouter, Vercel, and Cloudflare.
- Features an entirely new safeguard stack, achieving a top score of 62.4% on LatchBio's biosafety benchmark.
SpaceXAI has officially announced the release of Grok 4.7, a larger foundational artificial intelligence model that maintains the previous pricing structure of 2 dollars per million input tokens and 6 dollars per million output tokens. Users can deploy the model immediately as a hosted service through multiple integration channels, including the xAI API, Cursor, Grok Build, OpenRouter, Vercel, and Cloudflare.
In the launch comparison table pitting Grok 4.7 at xHigh effort against Grok 4.6 High, GPT-5.6 Sol Max, and Fable 5.1 Max, Grok 4.7 demonstrates improvements across every single benchmark row. The most substantial leap occurs on Terminal-Bench 4.0, surging from 20.3% to 38.0%. Furthermore, EEBench rose by 11 points to reach 64.0%, capturing the top score in the table, while Harvey's legal agent benchmark registered 19.6% compared to 6.7% for Fable 5.1 Max.
Stock photo for illustration only, not from the actual event
Despite these gains, Grok 4.7 does not lead across all domains. Fable 5.1 Max maintains the top spot in 4 out of 7 benchmarks, including a 57.9% Terminal-Bench score, while GPT-5.6 Sol Max holds the highest DeepSWE v1.1 result at 72.7%. Regarding cost dynamics, Fable 5.1 Max is approximately 5 times more expensive on input and 8.3 times higher on output, whereas GPT-5.6 Sol Max costs twice as much on input and about 3.3 times more on output, placing Grok 4.7 at the price-performance frontier on the CursorBench 4.0 cost chart.
On GDPval, which measures professional knowledge work, Grok 4.7 xHigh scored 1,695 Elo, improving upon the 1,605 Elo achieved by Grok 4.6 High. Fable 5.1 Max leads this category at 1,735 Elo, while GPT-6 Astra max recorded 1,542 Elo. SpaceXAI also noted that Grok 4.7 excels further in document creation and presentation generation. Alongside these updates, the model ships with an entirely new safeguard stack, which the company describes as its strongest system yet tested against refusals and jailbreak resistance, topping LatchBio's biosafety benchmark at 62.4%.
The introduction of Grok 4.7 highlights the ongoing industry trend of expanding model scale and reasoning capabilities while aggressively defending price-to-performance ratios. By holding the cost line at 2 and 6 dollars per million tokens while boosting coding and professional task benchmarks, SpaceXAI provides developers and enterprise builders with a highly competitive option for cost-sensitive scaling.
Concerning cybersecurity metrics, HackerBench v0.3—SpaceXAI's proprietary evaluation framework for risky and malicious cyber tasks—allowed 3.3% of risky dual-use prompts through. The company emphasizes that it rarely blocks legitimate security workflows. Selected cybersecurity partners are now being granted invite-only access to its red-teaming capabilities for defensive research initiatives.
For specialized deployment needs, Grok 4.7 Fast offers the same underlying model running on accelerated infrastructure, delivering twice the output speed at twice the standard price. Documentation specifies that this variant operates exclusively within Cursor and Grok Build, excluding the public xAI API and Grok Build's free tier. Additionally, a US regional endpoint located at https://us.api.x.ai/v1 routes inference requests domestically within the United States at a 10% premium, with recommendations to configure a prompt_cache_key for optimal cache retention.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment