SpaceXAI released Grok 4.7 with base prices unchanged at $2 per million input tokens and $6 per million output tokens, but its benchmark gains came with higher token consumption. On the Artificial Analysis Intelligence Index, it scored 46, up two points from Grok 4.6, while using about 81,000 output tokens per task versus 36,000. The evaluation compared Grok 4.7 at its xhigh reasoning setting with Grok 4.6 at high. The new model is available in Cursor, Grok Build and the Grok API.

The gains were stronger in coding and professional work. Paired with Grok Build, Grok 4.7 scored 56 on the Artificial Analysis Coding Agent Index, up from 47, with both models tested at xhigh. On AA-Briefcase, which tests professional assignments, it ranked behind only Anthropic models and just below Opus 5 at roughly half its cost per task. Producing example due diligence decks nevertheless cost about $8 with Grok 4.7 versus $4.40 with Grok 4.6, both at xhigh.

Results varied across other tests. Cognition made Grok 4.7 available in Devin and reported a 59.4% score on FrontierCode 1.1 Extended, but said it trailed Grok 4.6 overall because it took on too broad a scope on some engineering tasks. Vals AI initially ranked the model 24th on its index at 54.2%, below its predecessor’s 59.2%. After a software development kit update, Grok 4.7 rose to 10th.