Grok 5

xAI's next flagship — expected Q2 2026, reportedly 6 trillion parameters, trained on a gigawatt-scale supercluster. Here is what is confirmed and what is still rumour.

Grok 5 is the next flagship AI model from xAI, expected in Q2 2026. It is reported to have 6 trillion parameters and a Mixture-of-Experts architecture, trained on the Colossus 2 supercluster, and to be natively multimodal. Grok 5 has not yet launched, so no benchmarks exist.

Status: not yet released

This page tracks Grok 5 ahead of launch. Specifications below are reported or expected, not confirmed final figures — we update it as xAI publishes official details.

Grok 5 release date

Grok 5 is expected in Q2 2026. Elon Musk originally confirmed a Q1 2026 release, but Q1 came and went without a launch, and xAI's official X account now points to Q2 2026. No exact date has been confirmed — treat any specific day as speculation until xAI says otherwise.

What we know about Grok 5

MakerxAI
Expected releaseQ2 2026 (Q1 deadline missed)
ParametersReported 6 trillion — Mixture-of-Experts
Scale claimLargest publicly announced AI model to date
ModalitiesReported natively multimodal — text, image, audio, real-time video
Training hardwareColossus 2 supercluster, Memphis, Tennessee
BenchmarksNone — model not yet released

If the 6-trillion-parameter figure holds, Grok 5 would be the largest publicly announced AI model so far. The Mixture-of-Experts design means only a fraction of those parameters would activate per token, keeping inference cost workable. xAI also describes Grok 5 as built for real-time video alongside text, images and audio — a step beyond the static-media multimodality of most current models.

Trained on Colossus 2

Grok 5 was trained on Colossus 2, xAI's AI supercluster in Memphis, Tennessee — described as the world's first gigawatt-scale AI training cluster. The scale of the training infrastructure is xAI's central pitch for Grok 5: more compute than any previous xAI run, aimed at a decisive capability jump over the Grok 4 series.

Expected pricing

xAI has not announced Grok 5 pricing. What the trend suggests: xAI has been aggressive on price. Grok 4.1 Fast, at $0.20 per million input tokens, is currently among the cheapest frontier-class models available. On the consumer side, SuperGrok runs $30/month for the standard tier and $300/month for the 16-agent "Heavy" system. Competitive pricing is expected to carry over to Grok 5, though final numbers will only be clear at launch.

The current Grok lineup

While Grok 5 is pending, xAI's shipping models are worth knowing:

How Grok 5 may compare

Until Grok 5 ships with published benchmarks, any head-to-head is speculative. The released flagships to measure it against are below:

Visit xAI

Ad slot — AdSense in-article unit

Frequently asked questions

When is Grok 5 coming out?

Grok 5 is expected in Q2 2026. Elon Musk originally confirmed a Q1 2026 release, but Q1 passed without a launch and xAI's official channels now point to Q2 2026. No exact date has been confirmed.

How many parameters does Grok 5 have?

Grok 5 is reported to have 6 trillion parameters, using a Mixture-of-Experts architecture. If accurate, that would make it the largest publicly announced AI model to date.

What hardware was Grok 5 trained on?

Grok 5 was trained on Colossus 2, xAI's gigawatt-scale AI supercluster in Memphis, Tennessee — described as the world's first gigawatt-scale AI training cluster.

Will Grok 5 be multimodal?

Yes. Grok 5 is reported to be natively multimodal, designed from the ground up to handle text, images, audio and real-time video in a single architecture.

How much will Grok 5 cost?

xAI has not announced Grok 5 pricing. xAI has been aggressive on price — Grok 4.1 Fast costs $0.20 per million input tokens — and SuperGrok consumer plans run $30/month standard and $300/month for the 16-agent Heavy tier. Competitive pricing is expected to continue with Grok 5.

What is the latest Grok model available now?

As of May 2026 the current xAI flagship is Grok 4.20 Beta 2, which is competitive with — but does not clearly lead — GPT-5.4 and Claude Opus 4.7 across major benchmarks.