Grok 4.6 hits 61 on the Intelligence Index at the same $2/$6 — but the price doubles past 200K tokens

xAI shipped Grok 4.6 on August 12, 2026. Artificial Analysis scores it 61, up from Grok 4.5's 56 and tied with GPT-5.6 Sol, at unchanged $2/$6 per million tokens and a 500K context. Two details the launch post doesn't lead with: OpenRouter bills $4/$12 above 200K prompt tokens, and xAI's own card puts it at 26% on Terminal-Bench v3.0.

xAI released Grok 4.6 on August 12, 2026. (Source: xAI)

Key facts:

It scores 61 on the Artificial Analysis Intelligence Index, up from 56 for Grok 4.5. That ties GPT-5.6 Sol and trails Claude Opus 5 (63) and Claude Fable 5 (62).
Pricing is unchanged at $2 per million input tokens and $6 per million output. Cached input reads are $0.50/M.
Above 200,000 prompt tokens, OpenRouter bills $4/$12 with $1.00/M cache reads — a documented long-context tier, not a surcharge you find out about later.
The context window is 500,000 tokens, the same as Grok 4.5. The OpenRouter model id is x-ai/grok-4.6.
AA measures cost per Intelligence Index task at $0.84.
It is live today in Cursor, Grok Build and the xAI API, plus OpenRouter, Vercel and Cloudflare.

xAI's official bar chart of AA Intelligence Index scores: Fable 5 Max at 62, Grok 4.6 at 61 highlighted in orange, GPT-5.6 Sol Max at 61, and Grok 4.5 High at 56. — xAI's own comparison chart. The five-point jump over Grok 4.5 comes at identical list pricing. (Chart: xAI, 2026-08-12)

What this means if you’re building with Grok 4.6

1. The cost-per-intelligence position is the story, not the rank. On Artificial Analysis’s cost-versus-intelligence plot, Grok 4.6 sits at 61 for about $0.84 per task, while Claude Opus 5 reaches 63 at roughly $2.40 and Fable 5 62 at about $3. Same neighborhood of capability, a third of the spend. If you are running high-volume agent loops where a two-point index gap is inside your eval noise, that ratio is the argument.

Artificial Analysis scatter plot of Intelligence Index against cost per task on a log scale, showing Grok 4.6 at index 61 near $0.8 per task, Claude Opus 5 at 63 near $2.4, Claude Fable 5 at 62 near $3, GPT-5.6 Sol at 61 near $1.2, and Qwen3.8 Max at 58 near $1.7. — Grok 4.6 lands just outside the "most attractive quadrant" but well left of every model scoring above 60. (Chart: Artificial Analysis, 2026-08-12)

2. Token efficiency, not just token price, is where the savings come from. Artificial Analysis reports Grok 4.6 completing its agentic evaluations in roughly 53 turns and 0.5B input tokens on average, against roughly 103 turns and 2.0B input tokens for Claude Opus 5. Fewer turns means fewer round trips, less context re-sending, and a smaller share of requests crossing that 200K long-context threshold. Two models with the same sticker price can differ several-fold on the invoice for this reason alone.

3. Check Terminal-Bench before you swap it into a shell-driving agent. xAI’s own card lists Grok 4.6 at 26% on Terminal-Bench v3.0, against 34.6% for GPT-5.6 Sol and 34.1% for Fable 5 — the one row on the page where Grok is clearly last. Note the version: Artificial Analysis measures 88.4% on the older Terminal-Bench v2.1, so anyone quoting an 88% figure is citing a different, easier benchmark. If your agent spends its life in a terminal, run your own eval rather than either number.

4. Where it does lead is knowledge work. GDPVal-AA v2 puts it at 1753 Elo, second only to Claude Opus 5, and AA-Briefcase — an agentic knowledge-work benchmark — scores it 1577, ahead of Fable 5’s 1574 and behind only Opus 5’s 1715. Report generation, analysis, long-document work: that is the shape of task these two benchmarks reward.

5. Nothing about your integration changes. Same 500K context, same $2/$6 base, same endpoints as Grok 4.5. Swapping the model id is the whole migration, which makes this an unusually cheap A/B: run both against your eval set for a day and compare cost per completed task, not cost per token. If you drive Grok from the command line, our Grok Build guide and the CLI install walkthrough already cover the setup.

Sources: xAI — Grok 4.6 · Artificial Analysis — Grok 4.6 benchmarks and analysis · OpenRouter — x-ai/grok-4.6 pricing