xAI Ships Grok 4.7 at $2/$6 per Million Tokens, DeepSWE 71%
TL;DR
- xAI kept pricing flat at $2/$6 per million tokens despite a 40% parameter increase, pricing output at one-eighth of Claude Fable 5.1's rate.
- Artificial Analysis Intelligence Index scores Grok 4.7 at 46 versus 53 for Fable 5.1 and GPT-6, a capability gap independent data confirms.
- A 12-point Terminal-Bench discrepancy (38% xAI self-reported vs. 26% Artificial Analysis) signals first-party benchmark claims require independent corroboration before ops teams route workloads.
Grok 4.7, xAI's newest coding and knowledge-work model, ships at $2 per million input tokens and $6 per million output tokens, with a fast variant at double the price and double the output speed.
Benchmarks published by the company put the model at 46.3% on CursorBench 4.0, 71.0% on DeepSWE v1.1, 64.0% on EEBench, 56.7% on HealthBench Professional and 19.6% on the Harvey Legal Agent Benchmark. "Grok 4.7 is our most capable model for coding and knowledge work," the release states. "It works longer on difficult tasks, checks its own work more carefully, and comes with our best-calibrated safeguards to date."
On safety, xAI calls it "the strongest model we've tested on refusals and jailbreak resistance," and reports 62.4% on LatchBio's biosafety benchmark. Every score is self-reported; no third-party rerun is attached to the announcement.
The model is live today in Cursor and Grok Build, and through the Grok API, third-party coding harnesses, and model routers and cloud platforms. It caps a busy stretch for the company. Our xAI feed logged Grok Voice Transcribe 2.0 two days earlier.
What others are reporting
-
The Decoder Read →
Places the benchmark gap (Artificial Analysis: 46 vs. 53 for Fable 5.1) as the reason for bargain pricing, reframing xAI's cost advantage as a capability reflection.
Pricing sits at $2 per million input tokens and $6 per million output tokens. Those rates are closer to Chinese models than Western frontier models, probably for good reason.
-
DataCamp Read →
Most comprehensive specs breakdown: documents Terminal-Bench discrepancy explicitly, flags the 500K context window omission from official docs, and surfaces EEBench (64%) as a SpaceX-training strength.
Frontier-adjacent scores at $2/$6 per million tokens, an eighth of what Fable charges on output.
-
Decrypt Read →
Places the release within xAI's consistent pattern of trailing on capability while competing on price, citing second-place finishes on GDPval, AA-Briefcase, and CursorBench behind Fable 5.1.
a notable improvement over Grok 4.6 at the same price and speed
-
Cryptopolitan Read →
Sole source to disclose Cursor's SpaceX ownership as a benchmark-selection conflict; also documents Harvey Legal Agent scores (19.6% vs. Fable 5.1's 6.7%) as a genuine xAI strength.
Grok 4.7 scored 19.6% on the Harvey Legal Agent Benchmark. Fable 5.1 got 6.7%.
-
Android Headlines Read →
Developer-focused framing: highlights SpaceX engineering dataset integration and the 40% parameter expansion as practical value for large-scale batch coding workloads.
Keeping input costs at $2 per million tokens makes Grok 4.7 way more attractive for developers running massive batch coding scripts.
Shared on Bluesky by 1 AI expert
Originally reported by x.ai
Read the original article →Original headline: SpaceXAI Ships Grok 4.7 at $2/$6 per Million Tokens, Claims 71% on DeepSWE v1.1