Grok 4.6 matches GPT-5.6 Sol on the composite index at roughly a third of the price
SpaceXAI shipped Grok 4.6 on 12 August, five weeks after 4.5. Artificial Analysis puts it at 61 on its Intelligence Index — level with GPT-5.6 Sol, two points behind Claude Opus 5 — while pricing held flat at $2/$6 per million tokens against Sol's $5/$30.
The release cadence is the first thing worth noticing: five weeks between 4.5 and 4.6, and a five-point index gain in that window. Grok 4.3 to 4.6 is +23. Whatever SpaceXAI is doing to its post-training loop, it is compounding faster than the release notes suggest.
The composite number is 61, matching GPT-5.6 Sol and trailing Claude Opus 5 by two. On GDPVal-AA v2, a knowledge-work evaluation, Grok 4.6 posted an Elo of 1753 against 1728 for Sol and 1741 for Claude Fable 5. On CursorBench 3.2 it scored 69.9% against 67.2% for Sol and 70.5% for Fable 5. Context stayed at 500K. It was live in Cursor, Grok Build and the API the same afternoon.
Price is where the strategy shows. $2 in and $6 out per million tokens is roughly 60% below Sol's $5/$30, and SpaceXAI held it flat across a capability jump rather than harvesting the improvement. That is the same move Anthropic made this week when it cancelled a price increase it had already scheduled — two labs, same read on where the pressure is.
The thing not to conclude is that parity on a composite means parity in use. It does not, and the same scorecard says so.
SpaceXAI — Introducing Grok 4.6 → · Artificial Analysis — Grok 4.6 returns SpaceXAI to the intelligence frontier and leads on cost efficiency → · SiliconANGLE — SpaceXAI releases flagship Grok 4.6 model with advanced reasoning capabilities →