Editly

Grok 4.6: What We Know So Far

xAI's follow-up to Grok 4.5 has just landed, and the documentation has not caught up. Elon Musk confirmed Grok 4.6 on X in July 2026 and put it two weeks out with Grok 4.7 to follow; reporting describes a 1.5-trillion-parameter model on the same V9 foundation, improved through supervised fine-tuning and reinforcement learning rather than raw scale. The official model ID, context window, pricing, and benchmark table are still coming. This page tracks what is confirmed, what is reported, and what is still blank.

Start creating

What Is Grok 4.6?

A point release, not a new generation. Everything below is labeled by how firm it is: confirmed by xAI or Musk directly, reported by coverage ahead of the docs, or still unpublished.

Confirmed: The Model Exists and Ships Now

Musk confirmed Grok 4.6 in a reply on X on July 18, 2026, then posted on July 24 that Grok 4.6 was about two weeks out and Grok 4.7 about four. The rollout has now started. What has not appeared yet is a launch entry in xAI's release notes with the numbers developers need.

Reported: 1.5T Parameters on the V9 Foundation

Coverage describes Grok 4.6 as reusing Grok 4.5's 1.5-trillion-parameter V9 base rather than scaling up, with the gains coming from significantly better supervised fine-tuning and reinforcement learning. Treat the figure as reported until xAI publishes a model card.

Reported: Speed Held at Grok 4.5 Levels

Throughput and token efficiency are described as unchanged from Grok 4.5, which served in the range of 80 transactions per second. A same-generation refresh that keeps the serving envelope means no latency regression when you swap the model string.

Coming Soon: Model ID, Context Window, Pricing

No official API model ID, no published context window, and no separate price list at launch. Grok 4.5 is the working baseline: a 500,000-token context window at roughly $2 per million input tokens and $6 per million output. Expect 4.6 to land in that band until xAI says otherwise.

Coming Soon: The Benchmark Table

No official evaluation numbers have been released. For reference, Grok 4.5 posted 29.0 percent on SWE Marathon. Independent arenas and third-party evals will land before xAI's own table does — and those are the ones worth reading.

Next: Grok 4.7, Then Grok 5

Musk placed Grok 4.7 about two weeks behind 4.6, and reporting describes it as a 2.1-trillion-parameter step-up: better across the board, slightly slower to serve, with improved token efficiency. If you are planning a migration, 4.6 is a short-lived stop.

Grok 4.6 vs Grok 4.5

The Grok 4.5 column is documented; the Grok 4.6 column is what is reported or still pending. This table gets updated as xAI publishes real figures, so treat every pending row as a placeholder rather than a spec.

Grok 4.5
Grok 4.64.6
StatusDocumented flagship, generally availableJust released, official docs pending
Parameters1.5T on the V9 foundation1.5T, same foundation — reported
How it improvesBaselineBetter supervised fine-tuning and RL, not more scale
Context window500K tokensNot published yet
Price per 1M tokensAbout $2 in, about $6 outNot published yet
Serving speedAround 80 transactions per secondHeld at 4.5 levels — reported
Reasoning, tool calls, structured outputSupportedExpected to carry over
Official benchmarksPublished, including 29.0 on SWE MarathonComing soon
WeightsClosedClosed

Grok 4.6 Pricing: What to Expect

xAI has not posted a Grok 4.6 price list. Until it does, Grok 4.5's published rates are the honest planning baseline, and the table below shows where that sits against the open-weight alternative that shipped the same week.

Grok 4.64.6
Grok 4.5
DeepSeek V4 Pro 0813
Input per 1M tokensNot publishedAbout $2$0.435
Output per 1M tokensNot publishedAbout $6$0.87
Context windowNot published500K tokens1M tokens
Consumer accessGrok apps, grok.com, SuperGrok, X Premium+Grok apps, grok.com, SuperGrok, X Premium+DeepSeek app, free
Self-hostingNot availableNot availableAllowed, MIT weights

How to Get Grok 4.6

  1. 01

    Step 1: Check the Consumer Surfaces First

    xAI ships new Grok models to its own apps before the API docs are finished. Open the Grok app or grok.com and look at the model picker; SuperGrok and X Premium+ subscribers usually see a new flagship there first.

  2. 02

    Step 2: Watch the API Model List, Not the Blog

    In the xAI developer console, list available models programmatically rather than waiting for an announcement. A new model string appearing in that response is the earliest reliable signal that 4.6 is callable — release notes tend to trail it.

  3. 03

    Step 3: Keep Grok 4.5 in Production Until Numbers Land

    With no published context window, pricing, or benchmarks, treat 4.6 as an evaluation target. Put the model name behind a config value, run your own eval set against both, and switch only when 4.6 wins on your traffic. Grok 4.7 is reportedly weeks away, so avoid hardcoding anything.

  4. 04

    Step 4: Use It Where a Text Model Actually Helps

    Grok is text and reasoning. If the deliverable is a video or an image, use it to write the prompt — subject, camera move, lighting, beat-by-beat timing — and hand that text to a generation model. Editly renders it with Seedance, Veo 3, Hailuo, or Qwen Image.

Frequently Asked Questions about Grok 4.6

Have a Prompt? Render It Now

While Grok 4.6's specs firm up, the video side is ready today. Sign up for Editly and run Seedance, Veo 3, Hailuo, and Qwen Image on one credit balance — no per-model subscription.

Try Editly Free