SpaceXAI shipped Grok 4.6 on August 12, positioned as matching Fable 5 performance at a dramatic discount. On the Artificial Analysis Intelligence Index—a composite benchmark—Grok 4.6 posts a 61, level with GPT-5.6 Sol Max and one point behind Fable 5 Max at 62. The model is priced at $2/$6 per million input/output tokens, 60%+ below Claude Opus 5 and GPT-5.6 Sol. The parity claim, however, holds narrowly: on head-to-head benchmark counts, Grok 4.6 loses to Fable 5 Max on 7 of 10 shared tests.
On long-horizon agentic work (AA-Briefcase), Grok 4.6 achieves Fable 5 tier with an Elo of 1577 while using 53 turns and 500M input tokens on average—half the turns and a quarter of the tokens Claude Opus 5 (max) requires (~103 turns, ~2B tokens). This efficiency advantage compounds on per-task cost: Grok 4.6 costs $0.84 per task on the same benchmark, placing it on the cost-efficiency Pareto frontier. On pure intelligence metrics (GDPVal-AA v2 at 1753), Grok 4.6 leads both Fable 5 Max (1741) and Sol (1728).
For production architects and platform teams, the signal is economic: Grok 4.6 trades a marginal intelligence gap (wins 3/10 vs Fable, 6/9 vs Sol) for dramatically lower TCO on agentic and coding workloads. The cost advantage is largest on long-running tasks where token-burn matters more than per-token rate. Context window stays at 500K, but pricing jumps at the 200K-token boundary, creating a trade-off at scale.