After a series of timeline pushbacks since late July, Elon Musk’s SpaceXAI has officially rolled out Grok 4.7. The new release has a bigger architecture, tuned for complex coding, professional knowledge work, and multi-hour reasoning. Despite the overhaul, xAI keeps standard token pricing identical to Grok 4.6 at $2 per million input tokens and $6 per million output tokens, alongside a high-speed variant offering double the output speed at twice the token cost.
Under the hood, Grok 4.7 runs on a new base model packing 2.1 trillion parameters. The figure represents a 40% expansion over its predecessor’s 1.5 trillion. The training run relied on a longer reinforcement learning schedule using a harder mix of multi-hour problems.
To give the AI an advantage when solving hardware and physical system problems, xAI also supplemented its training dataset with internal engineering records from SpaceX. This includes Starlink satellite telemetry, manufacturing logs, and failure analysis data.
Benchmark performance across coding, legal, and agentic tasks
The extra scale and native harness integration show up clearly across technical benchmarks. On DeepSWE v1.1 at high effort, Grok 4.7 reached 71.0%, up from 65.2% on Grok 4.6. For coding workloads on CursorBench 4.0, it posted 46.3% accuracy, outperforming GPT-5.6 Sol Max (41.7%) while trailing Anthropic’s Claude Fable 5.1 (51.8%). On Terminal-Bench 4.0, it registered 38.0%.
The model also recorded notable bumps in specialized fields, scoring 64.0% on EEBench for electrical engineering and jumping to 19.6% on the Harvey Legal Agent Benchmark.
On enterprise agent tests managed by Artificial Analysis, Grok 4.7 logged 1657 Elo on AA-Briefcase (a 111-point gain over Grok 4.6) and hit 1695 Elo on GDPval. While it sits just behind Claude Fable 5.1 on both leaderboards, the score jump closes a significant gap in multi-hour professional workflows. Musk noted that future updates—Grok 4.8, Grok 4.9, and Grok 5—will target frontier leadership.
Redesigned safety stack and broad developer access
Safety guardrails saw a complete overhaul with this release. xAI introduced a new safeguard stack that topped LatchBio’s biosafety benchmark at 62.4%. On HackerBench v0.3—xAI’s internal test for cyber risk—the model blocked all but 3.3% of dangerous dual-use prompts without flagging legitimate security research. Select cybersecurity partners are also getting invite-only access to red-team tools for defense research.
Grok 4.7 is live without a waitlist across Cursor, Grok Build, third-party model routers, and the xAI API. It is also available natively in the Grok app, on X, and through Tesla dashboard voice assistants.
The post Grok 4.7 Brings Big Coding Upgrades to Challenge Claude AI at Unchanged Pricing appeared first on Android Headlines.
​Â
