Elon Musk’s xAI released Grok 4.7 on Monday after at least five public timeline delays since late July. The new model went live immediately in the Grok app, Cursor, Grok Build, and the xAI API, with no waitlist. Musk called it "a strong combination of intelligence, speed & low cost."
Grok 4.7 runs on 2.1 trillion parameters, a 40% increase over Grok 4.6’s 1.5 trillion. Pricing remains unchanged at $2 per million input tokens and $6 per million output tokens. That undercuts OpenAI’s GPT-5.6 Sol, which charges $4 for input and $20 for output, and dramatically undercuts Anthropic’s Fable 5.1 at $10 for input and $50 for output.
In benchmarks, Grok 4.7 delivered stronger results on some specialized tasks but still trailed the frontier pack in others. It scored 19.6% on the Harvey Legal Agent Benchmark, compared with 6.7% for Anthropic’s Fable 5.1 and 2.5% for OpenAI’s GPT-5.6 Sol. xAI said Grok 4.6 previously scored 15.8% on the same test. Grok 4.7 also beat Fable 5.1 on EEBench electrical engineering and edged past it on the DeepSWE coding test.
However, Anthropic’s Fable 5.1 retained leads on several broader measures. Grok 4.7 posted 1,695 Elo on GDPval, behind Fable 5.1’s 1,735 but ahead of GPT-6 Astra’s 1,542. On AA-Briefcase, Grok 4.7 scored 1,657 versus Fable 5.1’s 1,678. The widest gap appeared on Terminal-Bench 4.0, where Fable 5.1 reached 57.9% and Grok 4.7 reached 38.0%. On CursorBench 4.0, Grok 4.7 achieved about 46% at roughly $6 per task, while Fable 5.1 reached 51.8% at around $17 per task.
xAI included supplemental training data from SpaceX, such as Starlink satellite telemetry, manufacturing records, and engineering failure logs. The company said Grok 4.7 spends more time on hard problems and double-checks its answers more often than Grok 4.6. Musk also outlined future models: Grok 4.8 as a meaningful step up, Grok 4.9 in "Astra/Fable class," and Grok 5 as a possible frontier leader, though no release dates were given.
