SpaceXAI shipped Grok 4.6 today, scoring 61 on Artificial Analysis's nine-benchmark Intelligence Index, tied with OpenAI's GPT-5.6 Sol and one point behind Anthropic's Claude Fable 5. The pricing is the story: $2 per million input tokens and $6 per million output tokens below the 200K-prompt threshold, less than half what GPT-5.6 Sol costs through OpenAI's API for the same composite score.

The release also lands in Cursor and Grok Build on day one, with a 2x included-usage multiplier for the first week.

What makes this notable isn't the leaderboard placement. It's that, per MarkTechPost, "Grok 4.6 is not a larger base model." SpaceXAI's own release notes describe a post-training program on the Grok 4.5 foundation: a longer supplemental training run, curated model-generated data on reasoning and engineering concepts, an updated optimiser, SFT trajectories regenerated across reasoning-effort levels and agent harnesses, then RL across knowledge work, general coding, and web development. Five composite points on the Intelligence Index, extracted from the same weights class.

The coding numbers move in step. DeepSWE climbs from 54% on Grok 4.5 to 65.9%. CursorBench moves from 66.7% to 69.9%. The context window is 500,000 tokens. Knowledge cutoff is February 1, 2026. Above the 200K-prompt threshold, tokens re-price to $4 input and $12 output, with cached input at $1; below it, cached input runs $0.50. A "fast" variant costs twice standard.

VentureBeat's read on the release calls it "frontier-level intelligence, large improvements over the previous generation, stronger long-running agent behavior and relatively aggressive token economics." That last phrase is doing the work. The frontier tier through 2025 was defined by lockstep pricing between OpenAI and Anthropic, with challengers arriving cheaper but visibly behind on benchmarks. Grok 4.6 breaks that pairing. Same score as GPT-5.6 Sol, half the API cost, and a $30/month SuperGrok plan for the consumer surface.

The distribution posture reads as deliberate. Same-day availability through Cursor, Grok Build, OpenRouter, Vercel, and Cloudflare puts the model in front of the developer audience most sensitive to per-token economics, at the exact moment its price advantage is fresh. The 2x usage multiplier in the first week is a classic acquisition mechanic borrowed from consumer fintech: subsidize the trial window, let the switching cost accrue.

The structural read is that frontier parity is no longer a moat. When a post-training refresh on last generation's base can tie the flagship of the incumbent lab at half the price, the competitive question shifts from who trains the biggest model to who monetizes the smallest usable one. SpaceXAI just answered in public.

Grok 4.6 ties GPT-5.6 Sol at 61 on Artificial Analysis, holds the $2/$6 line

Sources