Some time ago, Musk and his SpaceXAI were actually regarded as lost in the AI market. However, the new Grok 4.6 is forcing critics to reconsider their views. This model competes with Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 Sol, but is much cheaper.
In the Artificial Intelligence Analysis Index, Grok 4.6 scored 61 points, which is the same as GPT-5.6 Sol and only 1 point lower than Claude Fable 5. The new SpaceXAI model also topped the GDPVal-AA v2, AA-Briefcase and Harvey LAB (Vals) tests. Meanwhile, Grok 4.6 costs just $2 per 1 million input tokens and $6 per 1 million output tokens, which is about half as much as other advanced models.
However, in the context of price competition, more serious competitors have emerged. Today DeepSeek launched the V4-Pro model, setting the price at $0.435 per 1 million input tokens and $0.87 per 1 million output tokens.
At the same time, SpaceX continues to actively increase its computing power. The company plans to launch around 2 GW of computing capacity by the end of 2026, and by the end of 2027 it hopes to be closer to the 10 GW mark rather than 5 GW. In this case, the total power and cooling load can reach 20GW.
The scale of the infrastructure is already impressive. As of May 2026, the Colossus 1 data center hosts more than 220,000 NVIDIA GPUs, including H100, H200 and approximately 30,000 GB200 accelerators. Colossus 2 gets even more powerful: it now has over 550,000 GPUs, including GB200 and GB300 accelerators.