xAI Launches Grok 4.7. It's Bigger, But Late to the AI Frontier Party
xAI on Monday launched Grok 4.7, an upgrade to its conversational AI that runs on 2.1 trillion parameters and is available immediately across xAI’s platforms and API. The company says the model improves on Grok 4.6’s reasoning and safety while keeping the same pricing, but independent benchmarks place it behind Anthropic and OpenAI’s leading models.
Why It Matters
The release marks another step in xAI’s strategy of offering lower-cost, widely available models rather than chasing top benchmark rankings, and it incorporates SpaceX engineering data aimed at improving reasoning about hardware and physical systems. How Grok 4.7 performs relative to competitors will influence xAI’s positioning in both consumer and specialized use cases such as in-car assistants and engineering support.
Key Facts
- Release date: Monday (date not specified in source excerpt) — launched after multiple delays since late July
- Model size: 2.1 trillion parameters (40% increase from Grok 4.6's 1.5 trillion)
- Availability: Live in the Grok app, Cursor, Grok Build, and the xAI API with no waitlist
- Training data: Supplemental training on SpaceX data including Starlink telemetry, manufacturing records, and engineering failure logs
- Pricing: $2 per million input tokens and $6 per million output tokens (same price as Grok 4.6)
Elon Musk’s xAI rolled out Grok 4.7 after several postponed launch estimates that began in late July. The company describes the new model as a meaningful upgrade over Grok 4.6, saying it spends more time on difficult problems, double-checks its own outputs more often, and includes its strongest safety guardrails to date. xAI also emphasized that Grok 4.7 is available immediately across its consumer and developer surfaces with no waitlist. Grok 4.7 increases model capacity to 2.1 trillion parameters, a 40% jump from the previous 1.5 trillion-parameter Grok 4.6. xAI added supplemental training material drawn from SpaceX — including Starlink satellite telemetry, manufacturing logs, and failure records — with the stated goal of improving the model’s ability to reason about hardware and physical systems beyond what is feasible using internet text alone. Independent benchmark results in the source materials show Grok 4.7 remains behind some rival models. On GDPval, which measures performance on economically valuable knowledge work with an Elo-style score, Grok 4.7 scored 1695 while Claude Fable 5.1 scored 1735. On AA-Briefcase, a multi-hour office-work benchmark, Grok 4.7 posted 1657 versus Fable 5.1’s 1678. For coding tasks evaluated by CursorBench 4.0, Grok 4.7 ranked mid-pack: it was costlier per task than some models such as GPT-6 Astra and Claude Sonnet 5 but still trailed Fable 5.1 across price points. Those results echo earlier patterns for xAI, which has previously released models that prioritized speed, personality, or cost over topping benchmark leaderboards. xAI’s price-per-token strategy — lower token costs compared with some competitors — underpins its argument that broad availability and affordability can outweigh absolute top-end capability for many everyday users. Musk has signaled future releases (Grok 4.8, 4.9, and Grok 5) as planned steps toward closing the capability gap, though no dates were provided in the source material.
Keep Reading
X will now tell users when governments have forced it to limit their posts
Meta’s AI agent has been blocked from using Amazon.com
Meta’s Muse is outpacing ChatGPT’s early mobile launch
