SpaceX launches Grok 4.7 with long-horizon processing

SpaceX has launched Grok 4.7, its most capable model for coding and professional knowledge work, keeping standard pricing at an affordable $2/$6 per million tokens. While the model shows clear benchmark gains over Grok 4.6, independent testing by Artificial Analysis reveals that high token consumption makes its actual cost per task ($3.74) higher than rivals like GPT-5.6 Sol ($1.99). The model features a larger base architecture, enhanced reinforcement learning, and a new safety stack.

Grok 4.7 model release

• ▪SpaceX, which absorbed Elon Musk's xAI, released Grok 4.7, its most capable large language model to date for coding and professional knowledge work, on September 21, 2026
• ▪Grok 4.7 was released less than a week after SpaceX introduced Grok Voice Transcribe 2.0, a text-to-speech model

Benchmark performance comparison

• ▪On the Artificial Analysis Intelligence Index v4.3.2, a combination of ten benchmarks, Grok 4.7 scored 46, placing it behind Claude Fable 5.1 and GPT-6, which both scored 53
• ▪Grok 4.7 scored 64.0% on EEBench and 71.0% on DeepSWE v1.1, representing increases from Grok 4.6's scores of 53.0% and 65.2% respectively
• ▪Grok 4.7 scored 46.3% on CursorBench 4.0, an improvement from the 40.4% scored by Grok 4.6
• ▪On Terminal-Bench 4.0, Grok 4.7 scored 26% at its highest reasoning setting, trailing GPT-6 Astra at 60% and Claude Fable 5.1 at 55%

Token pricing structure

• ▪Standard pricing for Grok 4.7 is set at $2 per million input tokens and $6 per million output tokens, matching the same $2 per million input token and $6 per million output token rates used for Grok 4.6
• ▪For inputs above 256K tokens up to the 500K maximum context window, Cursor charges $4 per million input and $12 per million output tokens for standard Grok 4.7, and $6/$18 for the Fast tier
• ▪SpaceX offers a fast variant of Grok 4.7, known as Grok 4.7 Fast, that processes prompts twice as fast for double the price, at $4 per million input tokens and $12 per million output tokens

Cost per task economics

• ▪Artificial Analysis calculated the cost per Artificial Analysis Intelligence Index task at $3.74 for Grok 4.7 xHigh and $2.73 for Grok 4.7 High, compared to $1.99 for GPT-5.6 Sol Max
• ▪Testing by Artificial Analysis shows Grok 4.7 xHigh used approximately 81,000 output tokens per Artificial Analysis Intelligence Index task, compared to 36,000 for Grok 4.6 High and 27,000 for GPT-6 Astra Max

Technical architecture improvements

• ▪Grok 4.7 is designed to integrate with SpaceX's Grok Bot harness, a collection of technical resources that allows the model to split complex work among multiple parallel AI agents
• ▪Grok 4.7 incorporates a new safeguard stack, with SpaceX reporting a 62.4% score on LatchBio's biosafety benchmark and a 3.3% leak rate on HackerBench v0.3
• ▪SpaceX built Grok 4.7 using a new, larger base model than its predecessor, Grok 4.6, and subjected Grok 4.7 to a longer reinforcement-learning run with more difficult tasks