xAI ships Grok 4.7 at unchanged prices, aimed at long coding jobs

xAI released Grok 4.7 on September 21 across Cursor, Grok Build, its API and third-party platforms, a few days later than Elon Musk first promised. Prices did not move: $2 in and $6 out per million tokens below 200,000 tokens, doubling above that — and crossing the line bills the entire request at the higher rate, which matters for the long agent runs this model is built for.

The upgrade is a bigger base model with more reinforcement learning rather than a new tier or a longer window; the context stays at 500,000 tokens, with effort levels up to “xhigh”. xAI's own numbers show the gains concentrated in multi-hour work: Terminal-Bench 4.0 from 20.3% to 38.0%, DeepSWE from 65.2% to 71.0%, CursorBench from 40.4% to 46.3%, HealthBench Professional from 48.5% to 56.7%. These are self-reported, and independent index scores put it level with an open-weight model released the same morning at a fifth of the price.

• Reported Released September 21 across Cursor, Grok Build and the API; 500,000-token context; effort levels up to “xhigh”. LLM Stats
• Claimed xAI’s own numbers: Terminal-Bench 4.0 from 20.3% to 38.0%, DeepSWE from 65.2% to 71.0%. LLM Stats
• Reported Crossing the 200k-token threshold bills the whole request at the higher $4 / $12 rate. LLM Stats