Grok 4.5 Release Date: July 8, 2026 — US and EU Confirmed

If you searched for the Grok 4.5 release date and got two different answers, you are not imagining it. xAI's own API release notes date the launch to 8 July 2026. The announcement page at x.ai/news/grok-4-5 carries a byline of 16 July 2026. Both are real; only one is the release date. This page settles which, adds the regional rollout dates almost nobody publishes, and answers the question that actually matters now that Grok 4.6 is out — whether Grok 4.5 is still worth pointing an API key at.

When was Grok 4.5 released?

8 July 2026. The primary source is xAI's own developer release notes, which log the launch under a dated July 8 entry:

"Grok 4.5, SpaceXAI's model for coding, agentic tasks, and knowledge work, is now available on the xAI API. Priced at $2 / 1M input tokens and $6 / 1M output tokens, with configurable reasoning effort (low, medium, or high; default high)."

Two independent records line up with that date. OpenRouter's public model catalogue records the x-ai/grok-4.5 entry as created at 15:05 UTC on 8 July 2026 (checkable yourself against the created field on OpenRouter's /api/v1/models endpoint). And trade press covered the model as shipped the following morning — The Decoder's 9 July 2026 write-up states plainly that "Grok 4.5 is available now through Grok Build, Cursor, and the xAI console." Artificial Analysis also lists 8 July 2026 as the release date on its Grok 4.5 model page.

Was Grok 4.5 released on July 8, 2026, or July 16?

If you have seen "July 8" quoted and wondered whether it was a rumour that stuck, it wasn't. July 8 is correct. The July 16 figure comes from one place: the publication date stamped on xAI's marketing announcement page. Its embedded metadata reads "datePublished":"2026-07-16T00:00:00Z", with an identical dateModified, and the visible byline says "Jul 16, 2026".

The gap is a publishing artefact, not a second launch. The July 8 release-note entry links forward to that same announcement page as its "announcement" reference, and the announcement's own copy is written in the present tense of launch day ("Grok 4.5 is available today in Grok Build, in Cursor on all plans, and from the SpaceXAI console"). The developer-facing log is dated on the day the model went live; the marketing page carries a later CMS timestamp.

The practical rule: cite 8 July 2026 as the release date, and treat 16 July as the date of xAI's written announcement. If you need a single defensible date for documentation or a changelog, use July 8 and reference the release notes.

Was Grok 4.5 available in the US at launch?

Yes — the United States had it from day one, 8 July 2026, on every surface xAI shipped it to. There was no staggered US rollout, and no waitlist.

The regional nuance runs the other way. Grok 4.5 is served exclusively from US clusters: xAI's model page lists the available regions as us-east-1 and us-west-2, and nothing else. That is still true today, and it is identical for Grok 4.6. So for a US developer, "is Grok 4.5 available in the US" has always been a non-question — the model is US-hosted by design. If you have data-residency requirements outside the United States, that region list is the constraint to check, not the launch date.

Europe was the region that actually waited. Grok 4.5 was not available to EU users at launch; The Decoder's July 9 coverage noted xAI "targeting a mid-July launch" for the EU, and the release notes confirm it landed on 17 July 2026 with the entry "Grok 4.5 is now available in the API console for EU users." That is a nine-day gap between US and EU access — which is almost certainly the real source of the conflicting dates floating around, since the EU announcement window and the July 16 page timestamp sit right on top of each other.

Where it shipped on day one:

• xAI API / SpaceXAI console — model string grok-4.5, priced per token.
• Grok Build — Grok 4.5 became the default model there at launch. Our Grok Build skills and connectors guide covers that surface in depth.
• Cursor — on all plans, with free Grok 4.5 usage offered for a limited period.

One claim worth not repeating: xAI's launch announcement does not mention grok.com, the X app, X Premium or SuperGrok tiers at all. Plenty of write-ups assert a consumer rollout on those surfaces a day later. We could not verify it against any xAI source, so we are not stating it.

Is Grok 4.5 still available now that Grok 4.6 has shipped?

Yes. As of 31 August 2026, grok-4.5 is still a first-class entry in xAI's model catalogue, alongside grok-4.6, grok-4.3 and the grok-4.20 family. There is no deprecation banner and no published retirement date on its model page. xAI has retired older models before — the docs still carry a "Model Retirement on May 15" migration guide — so a future sunset is plausible, but nothing is scheduled.

Two details suggest it is more than a courtesy listing. First, it still carries live aliases: grok-4.5-latest and, notably, grok-build-latest. Second, it keeps full production rate limits — 150 requests per second and 50 million tokens per minute, the same ceiling as Grok 4.6. (Neither model supports the Batch API.)

Here is the confirmed specification, taken from the model page rather than launch marketing:

Worth correcting a widely-copied error while we are here: Grok 4.5's reasoning effort is configurable across low, medium and high, defaulting to high. It is not a locked always-on setting, and the levels are not just low/high. That comes straight from the July 8 release note.

Note also that the context window shrank relative to Grok 4.3, which still offers 1M tokens. If long-context is your binding constraint rather than reasoning quality, Grok 4.3 remains the better fit in the family — and it is cheaper, at $1.25/$2.50 per million.

How much does Grok 4.5 cost in 2026?

Pricing has not moved since launch. The one thing to watch is the context-tier cliff: cross 200,000 prompt tokens and the entire request bills at the higher rate.

Against the wider market — prices below read live from OpenRouter's catalogue on 31 August 2026 — Grok 4.5 still sits in the cheap half of the near-frontier tier, but its price advantage has narrowed sharply since July:

The single most useful line here is cached input. Grok 4.5 reads cached tokens at $0.30 per million; Grok 4.6 raised that to $0.50, a 67% increase. For a pipeline that replays a large fixed prompt — a long system prompt, a repository map, a document corpus — cache reads dominate the bill, and Grok 4.5 is genuinely the cheaper model. That is the clearest surviving reason to stay on it. For the full breakdown of what changed on the newer model, see our Grok 4.6 pricing and API costs guide.

How good is Grok 4.5, and how far has it slipped?

At launch, xAI's own comparison charts put Grok 4.5 at or near the top of several agentic coding benchmarks. Its figures, with competitor numbers drawn from those vendors' published system cards:

The honest read even at launch was that Grok 4.5 mostly placed third, with one outright win — SWE Marathon, a long-horizon agentic benchmark, where it beat both Opus 4.8 and Fable 5. That result is the most interesting thing on the chart and the one most write-ups skipped.

The efficiency claim held up better than the raw scores. xAI measured Grok 4.5 resolving SWE Bench Pro tasks with an average of 15,954 output tokens against Opus 4.8's 67,020 — 4.2× fewer — and served it at a claimed 80 tokens per second. Fewer tokens per finished task, multiplied by a low per-token price, is where the savings actually appeared.

Seven weeks on, the leaderboard has moved. On Artificial Analysis's Intelligence Index today, Grok 4.5 (high) scores 56 and ranks #21 of 187 models. At launch it sat in the top handful. It has not got worse; the field simply filled in beneath it, and xAI's own Grok 4.6 chart uses the same 56 figure as the baseline it improved on. Independently measured today, Grok 4.5 runs at 54.2 tokens per second with a 14.51-second time to first token, at roughly $0.43 per Intelligence Index task.

xAI also published some training detail worth knowing, since almost every secondary summary invents its own version: Grok 4.5 was trained across tens of thousands of NVIDIA GB300 GPUs in xAI's Memphis data centres, on datasets spanning coding, science, engineering and maths, and was "trained alongside Cursor." xAI has not published a parameter count, an architecture description, or a system card. The "1.5 trillion parameter mixture-of-experts" figure that circulates is not sourced to xAI and should not be repeated.

Should you still use Grok 4.5, or move to Grok 4.6?

Grok 4.6 is the better model on capability, and it is not close on the composite score. But it is not a free upgrade — it is slower to start and costs more per finished task. Both columns below are Artificial Analysis's own measurements, taken on the same harness on 31 August 2026:

The decision rule is genuinely simple:

• Stay on Grok 4.5 if your workload is latency-sensitive or cache-heavy. A 14.5-second time to first token versus 43.8 seconds is the difference between a usable interactive tool and one users abandon, and at $0.43 versus $0.94 per task you are paying 2.2× less for work Grok 4.5 can already complete.
• Move to Grok 4.6 for hard single-shot code, long-running agents, and anything where a failed task costs more than a slower one. xAI built 4.6 specifically around "long-running agents", and the five-point index gap is real.
• Look at Grok 4.3 if you need the 1M-token context window or the lowest price in the family.

For the head-to-head detail, we have a dedicated Grok 4.6 vs Grok 4.5 comparison and a full Grok 4.6 launch guide. If you are choosing a model to sit behind a coding agent rather than a chat box, our AI coding agents guide covers how these plug into day-to-day workflows.

FAQ

When was Grok 4.5 released?

Grok 4.5 was released on 8 July 2026, when xAI (SpaceXAI) made it available on the xAI API, in Grok Build, and in Cursor on all plans. The date is recorded in xAI's own API release notes and matches OpenRouter's catalogue entry, created at 15:05 UTC that day. EU API access followed on 17 July 2026.

Was Grok 4.5 released on July 8, 2026?

Yes. July 8, 2026 is the correct release date, confirmed by xAI's dated developer release notes. The competing "July 16" date comes from the publication timestamp on xAI's marketing announcement page, which was posted eight days after the model went live. If you need one date, use 8 July 2026.

Is Grok 4.5 available in the US?

Yes, and it was from day one on 8 July 2026 with no staggered US rollout. In fact the API is served only from US regions — xAI lists us-east-1 and us-west-2 as the sole clusters for Grok 4.5. The region to check is not the US but everywhere else; EU access arrived on 17 July 2026.

Is Grok 4.5 still available?

Yes. As of 31 August 2026, grok-4.5 remains in xAI's model catalogue with no published retirement date, full production rate limits of 150 requests per second and 50 million tokens per minute, and live aliases grok-4.5-latest and grok-build-latest. It is the previous generation, not a withdrawn one.

Should I use Grok 4.5 or Grok 4.6?

Use Grok 4.6 for hard code and long-running agents — it scores 61 on Artificial Analysis's Intelligence Index against Grok 4.5's 56. Use Grok 4.5 when latency or cache cost matters: it starts responding in 14.51 seconds versus 43.82, costs $0.43 per index task versus $0.94, and reads cached input at $0.30 per million versus $0.50.

How much does Grok 4.5 cost?

Grok 4.5 costs $2.00 per million input tokens, $0.30 per million cached input tokens, and $6.00 per million output tokens for requests under 200,000 prompt tokens. Above that threshold the whole request bills at the higher tier: $4.00 input, $0.60 cached, $12.00 output per million. Pricing has not changed since launch.

How big is Grok 4.5's context window?

500,000 tokens, with text and image input and text-only output. That is smaller than Grok 4.3's 1,000,000-token window, so if very long context is your constraint, the older model is still the better choice in the family. Grok 4.6 kept the same 500,000-token limit.

Is there a Grok 5?

Not as of 31 August 2026. Grok 4.6, released on 12 August 2026, is xAI's current flagship and the most recent entry in the API release notes. Grok 4.5 sits one generation behind it, with Grok 4.3 and the Grok 4.20 family still listed below both.

If you only need one line from this page: Grok 4.5 shipped on 8 July 2026, the US had it immediately, the EU got it on 17 July, and it is still running. It is no longer the model to reach for by default — Grok 4.6 is — but it remains the cheaper and much faster-starting option, and for cache-heavy or interactive workloads that is a real reason to leave your model string alone.

Building on frontier models and need engineers who already work this way? Codersera helps you hire vetted remote developers who ship with AI coding agents in the loop.