Grok 5 release date: not announced, and here is what that means.

Quick answer: what is actually known.

Three things are solid. First, Grok 5 is real as a project: Musk and xAI have discussed it publicly since 2025, including Musk's much-quoted remark that his estimate of Grok 5's probability of achieving AGI was "10% and rising", a statement about ambition rather than a spec sheet. Second, the timeline has slipped: targets Musk floated publicly, first late 2025, then early 2026, have passed without a release, and tracking coverage through mid-2026 records the slippage rather than a new committed date. Third, xAI's shipped 2026 flagship is Grok 4.5: announced June 28, 2026, released broadly July 8, 2026, positioned for coding and agentic work. Everything else in circulation, parameter counts, training-cluster details, launch windows, is reporting and rumor of varying quality, covered in its own section below so it does not blend into the verified record.

The announced record, dated.

Musk began talking about Grok 5 while Grok 4 was still current, with training-start claims in late 2025 and an initial release ambition of "end of year" that moved to early 2026 in later statements. Through the first half of 2026, xAI's public energy went to the Grok 4 line instead: point releases, the fast tier, and then Grok 4.5's June announcement and July release as the company's first model built specifically for coding and agent workflows, per its launch materials. That substitution is the most informative fact in the record. Labs ship what is ready; a strong point release in July is a statement about what was ready in July. It says nothing certain about Grok 5's internal state, but it does mean the burden of proof for any near-term Grok 5 date sits with whoever claims one, because the vendor has not. For readers keeping score of xAI's public pattern: announcements have historically come from Musk's X account and xAI's blog with short lead times, days to weeks between announcement and availability, so when Grok 5 does move, the gap between rumor and endpoint will likely be short.

The rumored record, labeled as such.

A note on labeling honesty: some mid-2026 coverage has mislabeled other model releases as Grok 5 sightings, which is how "Grok 5 is out" rumors briefly flare. The check that cannot be fooled is the models endpoint: if no grok-5 id appears where you can send tokens, it is not released, whatever a headline says.

• Scale: reports across 2026 describe Grok 5 as a roughly 6-trillion-parameter Mixture-of-Experts. No xAI specification confirms a parameter count, and pre-release parameter rumors have a poor accuracy record industry-wide. Speculation.
• Infrastructure: reporting ties training to xAI's Colossus 2 buildout, with gigawatt-scale framing in some coverage. Directionally plausible given xAI's public datacenter announcements, unconfirmed in the specifics. Speculation.
• Timing: mid-2026 tracker coverage converged on second-half-2026 guesses, with API access sometimes rumored to trail the consumer launch. These are observers extrapolating from slipped targets, not xAI commitments. Speculation.
• Capability: the AGI-probability remark gets quoted as if it were a benchmark claim. It is an aspiration statement from an executive famous for aggressive timelines, and it predates any measurable artifact. Not evidence.

What xAI actually shipped instead: Grok 4.5.

Grok 4.5 is the concrete answer to "what can I use from xAI right now". Announced June 28, 2026 and released broadly on July 8 per launch coverage, it targets coding and agentic work, carries a 500K-token context window as served in the catalog here, and launched with an efficiency story: coverage highlighted per-task costs landing well under the Western flagships it competes against, plus a steep cached-input discount on xAI's own platform. EU availability lagged the launch by design, with mid-July targeted, a regional detail worth knowing if your traffic originates there. For anyone who searched for Grok 5 because they want the strongest available xAI model with an API: that is grok-4.5 today, and it is served here without an xAI console account.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["APISROUTER_API_KEY"],
    base_url="https://api.apisrouter.com/v1",
)

resp = client.chat.completions.create(
    model="grok-4.5",   # released July 8, 2026; Grok 5 remains unreleased
    messages=[{"role": "user", "content": "Refactor this function for testability and explain why: ..."}],
    max_tokens=4000,
)
print(resp.choices[0].message.content)

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

How to be positioned for Grok 5 without waiting on it.

The teams that absorb model launches painlessly all share one architectural habit: model ids live in config, not code, and every model call goes through one endpoint that serves multiple vendors. When a grok-5 id eventually appears, that setup turns launch day into an A/B run, the new id against grok-4.5 and against the non-xAI frontier on your own evaluation set, rather than an integration project. The second habit is evaluative rather than architectural: keep a standing evaluation set of your real tasks, and rerun it when candidates appear. Launch-week benchmarks for every 2026 frontier model have diverged from month-three leaderboard standings, and a model that wins your rubric on your traffic is worth more than any headline number. Both habits cost nothing to adopt now, and they convert the question this page cannot answer, when Grok 5 ships, into a question that does not matter much, because whenever it ships, your switch is a string.

Why Grok release-date searches go wrong.

The failure pattern is specific to how xAI communicates. Ambitious public targets from Musk arrive months before engineering reality, get reported as dates, then slip without a formal retraction, leaving stale "Grok 5 coming this quarter" coverage indexed forever. Meanwhile actual releases move fast and casually, a post, a rollout, sometimes API access trailing consumer access, so the accurate signal is short-lived and the inaccurate signal is durable. Add the mislabeling problem, other releases briefly reported as Grok 5, and search results become an archaeology dig sorted by confidence rather than by date. The definitive check is the same as for every family on this site: list what your endpoint serves. The moment a grok-5 id answers requests somewhere you can authenticate, it is released for you; until then, every date you read is a guess wearing a headline.

curl -s https://api.apisrouter.com/v1/models \
  -H "Authorization: Bearer $APISROUTER_API_KEY" | grep grok

FAQ