PALO ALTO, Calif.: SpaceXAI releases Grok 4.6, a new flagship model the company says matches OpenAI’s GPT-5.6 Sol on the Artificial Analysis Intelligence Index.
The model launches with pricing unchanged from its predecessor. Input costs $2 per million tokens and output costs $6. A faster variant is available at double those rates. Access runs through Cursor, Grok Build, the company’s API, and partners including OpenRouter.
Grok 4.6 is optimized for long-running agents, coding, and multi-step interactive tasks. The company reports gains over Grok 4.5 across coding and knowledge benchmarks. It attributes the improvement to extended training runs, higher-quality engineering data, and improved optimizers.
The claimed benchmark parity marks a significant jump in a short period. Grok 4.5 launched on July 8 ranked fourth overall on the same index, with a score of 54. That model was positioned primarily on cost efficiency rather than capability leadership.
The pricing gap with rivals remains wide. OpenAI charges $5 per million input tokens and $30 for output on GPT-5.6 Sol. Grok 4.6 undercuts those rates by more than half on input and 80% on output.
The benchmark claims come from SpaceXAI’s own reporting and the Artificial Analysis index. Independent task-level evaluations typically take several weeks to accumulate after a release.
The launch continues a rapid release cadence at the lab, which SpaceX absorbed before its June initial public offering. SpaceX’s public filings disclosed that the AI unit lost $2.4 billion in the first quarter.
Grok 4.6 arrives during a crowded stretch for model releases. Meta shipped Muse Glimmer on August 10, and OpenAI launched a cybersecurity-focused model the same week.
Previous
