SpaceXAI's new Grok 4.6 scores within striking distance of GPT-5.6 Sol at a fraction of what the company still charges SuperGrok Heavy subscribers for last quarter's model.
Grok 4.6 went live on August 12, 2026. It landed across the xAI API, the Grok app, Cursor, OpenRouter, Vercel and Cloudflare, all at once. It scores 61 on the Artificial Analysis Intelligence Index, tying OpenAI's GPT-5.6 Sol and landing five points above Grok 4.5, according to Artificial Analysis's own benchmark writeup. Anthropic's Claude Opus 5 and Fable 5 still sit above it. But Grok 4.6 now beats Moonshot's open weight Kimi K3 outright.
Here's the part that actually matters for anyone paying for this stuff: the model costs $2 per million input tokens, $0.50 for cached tokens, and $6 per million output tokens, for any prompt under 200,000 tokens. Cross that threshold and the price doubles to $4 and $12. That's the same sticker Grok 4.5 carried. Nothing changed on paper. But Artificial Analysis's cost-per-task numbers tell a different story: running its standard benchmark suite costs $0.837 on Grok 4.6 against $0.360 on Grok 4.5, a 2.3x jump driven by the new model generating roughly 47% more output tokens to get to its answers.
Context capacity jumped too. Grok 4.6 handles a 500,000-token window and adds image input, function calling and configurable reasoning effort, aimed squarely at coding agents and long-running tasks. On agentic benchmarks the gains are real. DeepSWE climbs 11.9 percentage points over Grok 4.5, Terminal-Bench rises 10.3 points, and APEX-Agents gains 10.4 points. Turns matter more than points, though. Artificial Analysis found Grok 4.6 resolves tasks in about 53 turns, using roughly 0.5 billion input tokens on average. Claude Opus 5, running at max settings, needs about 103 turns and 2 billion input tokens to get there. Fewer turns. A quarter of the input tokens. That's a cost advantage that sits well outside the per-token price sheet.
Paying more for less #
Grok 4.6 now ships inside the $30-a-month SuperGrok plan. SuperGrok Heavy, the $300-a-month tier, remains the only confirmed consumer route to Grok 4.5. Frankly, that's a strange place for SpaceXAI to land: the cheaper plan now carries the newer, higher-scoring model, while the tier that costs ten times as much is stuck pointing subscribers at last generation's release. Companies usually reserve their priciest consumer tier for the best thing they've shipped. SpaceXAI has, at least for now, inverted that. xAI is shipping Grok 4.6 and 4.7 back to back in a release cadence no frontier lab has matched
xAI confirmed a 2-trillion-parameter Grok 4.6 on July 18, 2026 and has a larger Grok 4.7 already weeks behind it, a release cadence no other frontier lab has publicly matched. With Grok 4.5 already priced over 60 percent below GPT-5.5 and Opus 4.8, xAI is applying simultaneous pricing and pace pressure on OpenAI and Anthropic while explicitly...
The comparison that actually moves the market is against OpenAI and Google. GPT-5.6 Sol trades blows with Grok 4.6 on the Intelligence Index, but neither OpenAI nor Google has published a token price that beats $2 in, $6 out for a frontier-class model at this performance tier. Gemini 3.7's pricing runs higher on output tokens for comparable context lengths. That gap is what SpaceXAI is selling developers on: a model that ties the field on raw intelligence while asking less per token than what OpenAI or Anthropic charge for their top-line releases. That holds even after accounting for Grok 4.6's heavier token usage per task.
Rockets and reasoning models #
All of this lands against a much bigger backdrop. SpaceX absorbed xAI in an all-stock merger that closed on February 2, 2026, valuing the combined company near $1.25 trillion. The renamed SpaceXAI then listed on Nasdaq under the ticker SPCX in what became the largest IPO on record, raising roughly $75 billion. The corporate rebrand landed July 6, 2026, moving the company's X account from @xai to @SpaceXAI. Grok itself kept its name. That didn't change. Elon Musk's holding company now runs rockets and reasoning models under one roof, and Grok 4.6 is the first flagship release to ship entirely under the SpaceXAI banner.
None of that history changes what a developer actually pays this week. A team building an agent on Cursor or OpenRouter gets a model that beats last quarter's Grok on coding and terminal benchmarks, at the same headline rate xAI charged before the merger. The $300 Heavy tier, for the moment, is selling access to the model that just got outclassed by the plan one-tenth its price. That's the trade, for now.
Also read: OpenAI's New Ultrafast Mode Runs GPT-5.6 Sol on Cerebras Chips, Not Nvidia • AI Agent Approval Fatigue Is Quietly Undermining Startup Safety • California Approves Waymo's Biggest Robotaxi Expansion Across 18 Counties