What happened
On August 12, 2026, SpaceXAI released Grok 4.6, a frontier-tier language model that scores 61 on the Artificial Analysis Intelligence Index—tying OpenAI's GPT-5.6 Sol and sitting just behind Claude Opus 5 (63) and Fable 5 (62). The release comes with aggressive pricing: $2 per million input tokens and $6 per million output, roughly 60% cheaper than Anthropic's Opus 5 ($5/$25) and GPT-5.6 Sol ($5/$30).
The numbers
Grok 4.6 jumped 5 points over its predecessor Grok 4.5 on the Intelligence Index. On coding benchmarks, it posted a 65.9% on DeepSWE (up from 54%) and 57.5% on APEX-Agents (from 47.1%). The model supports a 500,000-token context window and is available immediately through the xAI API, Cursor, OpenRouter, Vercel, and Cloudflare. Elon Musk called it "objectively No. 1 when considering intelligence, speed & cost" and said Grok 4.7 should be ready in 3 to 4 weeks.
Why it matters
A year ago, calling Grok a frontier model would have been a joke. Today, it's a pricing problem for OpenAI and Anthropic. The gap between performance and cost is closing faster than expected, and if Grok 4.7 delivers on Musk's promise to "exceed all current models," the frontier may have to choose between lowering prices or losing developer share. For teams building agentic workflows, the economics just changed—but the dependency didn't.