Grok 4.6: 61 Points on the AI Index, More Than 60% Cheaper Than Claude Opus 5
Agentic AISpaceXAI’s Grok 4.6 just posted a 61-point score on the Artificial Analysis Intelligence Index — matching OpenAI’s GPT-5.6 Sol — while pricing its API at more than 60% below comparable models. If the numbers hold under real workload conditions, this is the pricing compression event the enterprise AI market has been watching for.
The Number
61 points on the Artificial Analysis Intelligence Index, tying GPT-5.6 Sol. Only Claude Opus 5 (63 points) and Claude Fable 5 (62 points) rank higher. Grok 4.6 represents a five-point improvement over Grok 4.5 in a single generational jump.
How the Math Works
The pricing gap is the story. Grok 4.6 is priced at $2 per million input tokens and $6 per million output tokens. Claude Opus 5 runs $5/$25. GPT-5.6 Sol is $5/$30. At enterprise query volumes, running equivalent workloads through Grok 4.6 costs roughly one-third of the Anthropic or OpenAI alternatives.
On the GDPval-AA v2 benchmark for real-world knowledge work, Grok 4.6 achieves an Elo score of 1,753, placing second behind Claude Opus 5. The model completes complex tasks in approximately 53 steps — roughly half the 103 steps Claude Opus 5 uses — which is relevant for agentic workflows where step count directly influences latency and cost accumulation.
Access is available via API, Cursor, Grok Build, and partners including OpenRouter, Vercel, and Cloudflare. Early users receive doubled usage quotas for one week.
What It Signals
The Grok 4.6 release continues a pattern of frontier-grade models arriving at sub-frontier prices. Earlier this year, DeepSeek V4 Pro established that near-top performance was achievable at dramatically lower token costs. Grok 4.6 confirms the pattern holds in the US market with a model available through enterprise integration channels teams are already using.
For procurement decisions, the benchmark convergence matters as much as the price. When models at the 61-62 point band score within statistical noise of each other on knowledge work benchmarks, token economics become the primary differentiator. The step-efficiency advantage — 53 steps versus 103 for comparable agentic tasks — compounds the cost benefit further when running multi-step workflows.
This also shifts the benchmark narrative. For much of 2026, Claude Opus 5 has held the leadership position teams benchmarked against. Grok 4.6’s 61-point score makes the frontier effectively a three-way tie between products at very different price points. Teams that haven’t re-evaluated model mix since Q1 should. The xAI Grok price compression pattern observed with Grok 4.3 is accelerating, not stabilizing.
The So What
Grok 4.6 is the clearest argument to date that model selection at the frontier should begin with economics, not benchmarks alone. At 61 points and $2/$6 per million tokens, the performance-per-dollar ratio is the strongest currently available for enterprise agentic workloads. Teams locked into higher-cost alternatives for reasons that made sense six months ago — ecosystem lock-in, integration maturity, safety controls — should evaluate whether those reasons still outweigh a 60-70% token cost reduction at equivalent measured performance. The model is API-accessible now.
Source: The Decoder
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →