xAI AI Updates: September 22, 2026

1. Grok 4.7 Ships Cheap, and the Benchmarks Explain Why

xAI released Grok 4.7 as its frontier model for coding, agentic tasks, and knowledge work, available on the xAI API as grok-4.7. It carries a 500K token context window, accepts text and images, and prices at $2 input, $0.50 cached, and $6 output per 1M tokens below 200K prompt tokens, doubling to $4/$1/$12 above that threshold. Reasoning effort is selectable across low, medium, high, and xhigh. Source

2. On Artificial Analysis It Scores 46 Against 53 for Claude Fable 5.1 and GPT-6

Artificial Analysis places Grok 4.7 at 46 points on its Intelligence Index, mid-pack and well behind Claude Fable 5.1 and GPT-6 at 53 each, with The Decoder reporting the gap widening further on agentic coding. The pricing is the compensating argument rather than an incidental detail. For workloads where a seven-point index gap is tolerable, the cost per token is the reason to look. Source

3. Grok Voice Transcribe 2.0 Is Available, With 1.0 Still the Default

xAI also shipped grok-voice-transcribe-2.0 alongside the model release, giving developers a choice between versions 1.0 and 2.0. Version 1.0 remains the default, so the upgrade is opt-in and existing transcription pipelines are unaffected until they pin the new identifier. Source