Grok 4.6 Reached the Frontier. It Costs a Fifth as Much.

Most marketers use ChatGPT to do general research and then call it an AI strategy. The ones outperforming them are deploying specialized agents built for specific jobs.

We put together 22 plug-and-play ChatGPT marketing agents that handle the work eating your week, each with built-in instructions and structured outputs ready to go in under 5 minutes.

Subscribe to Marketing Against the Grain and get all 22 free.

Inside you'll find:

Your competitors are already running agents like these.

Get 22 ChatGPT Marketing Agents free when you subscribe to Marketing Against the Grain today.

Get The Guide

3 things that happened while you were busy

1.  Grok 4.6 tied OpenAI's best model on the independent index.

SpaceXAI released Grok 4.6 on August 12, and it scores 61 on the third-party Artificial Analysis Intelligence Index, tying GPT-5.6 Sol Max and passing Kimi K3. Only Claude Opus 5 at 63 and Fable 5 at 62 score higher. That is a five-point jump from Grok 4.5, the model we covered exactly a month ago, and this time the number comes from an outside evaluator rather than a vendor chart.

2.  The price did not move. That is the actual story.

Grok 4.6 stays at $2 per million input tokens and $6 output, roughly half the sticker price of comparable frontier models, with one analysis putting it around 80% cheaper on input and 88% on output than Fable 5 Max. Same money as last month, frontier-tier scores this month. The 500,000-token context window and a new xhigh reasoning level came along for free.

3.  It got there without a bigger model.

The engineering detail worth knowing: this is a post-training upgrade, not a larger base model. SpaceXAI held the foundation constant and spent the gains on extended supplemental training, regenerated fine-tuning trajectories, and reinforcement learning in agentic environments. Proof that the cheap wins in 2026 are increasingly in training technique, not raw scale, which is the same lesson Bridgewater taught us in July.

Four labs are now within two points of each other. The tiebreakers are elsewhere.

The compression.  Look at the top of the index: Opus 5 at 63, Fable 5 at 62, and GPT-5.6 Sol Max tied with Grok 4.6 at 61. Four labs inside a two-point band. As VentureBeat put it, Grok 4.6 does not establish an uncontested performance lead; it offers frontier-level intelligence with aggressive token economics instead. When capability converges, price and reliability become the product.

Where it actually loses.  Read past the headline score and the picture sharpens. On coding, the rows engineering teams care most about, Grok 4.6 trails: DeepSWE at 65.9% against GPT-5.6 Sol Max's 73%, and Terminal-Bench at 26%, last of the four models listed. Where it shines is long-running agent work, ranking second on GDPval-AA v2 with an Elo of 1,753, completing complex tasks in about 53 steps. Strong agent, weaker coder. Pick accordingly.

The part benchmarks cannot score.  VentureBeat raised the point most launch coverage skipped: enterprise buyers rarely evaluate a model separately from its vendor and product history, and Grok's earlier deployments produced several brand-safety incidents. Nothing in this release repeats them, but organizations that cannot afford their AI supplier to become a news story will weigh that history. Reliability of the vendor is now part of the spec sheet.

The takeaway.  The release cadence is the thing to watch: reports point to a much larger Grok 4.7 within weeks and Grok 5 before year end, while DeepSeek shipped V4-Pro this week at $0.435 per million input tokens. Nobody should be signing a long commitment to any single model right now. Build so you can swap providers in an afternoon, and let them keep competing for your bill.

What people are actually watching and sharing

The 22-minute run.  A tech creator shared a demo where Grok 4.6 worked autonomously from one prompt for 22 minutes, producing a project with custom shaders, a minimap, and a time-change feature. Sustained autonomy without drifting is exactly what the agentic benchmarks were measuring.

Free double usage.  SpaceXAI is including 2x usage inside Cursor and Grok Build for the first week. If you were going to test it, this week costs you half as much as next week.

DeepSeek's counterpunch.  On the same day, DeepSeek released V4-Pro at $0.435 input and $0.87 output per million tokens, roughly a fifth of Grok 4.6's price. The floor keeps dropping while the ceiling keeps rising, which is a strange and excellent time to be a buyer.

Same brand, new parent.  If the name keeps confusing you: SpaceX acquired xAI in February, so the models now sit under a different corporate structure while keeping the Grok consumer brand. Same team, new letterhead.

Build a personal benchmark before the next model drops

Four models now sit within two points of each other, and two more ship within weeks. Public benchmarks cannot tell you which one is better at your work, and switching on vibes wastes money. Build your own scoring suite once, then every new launch becomes a fifteen-minute test instead of a guess.