Skip to main content
← All news
AI News

Grok 4.5 gets a Thursday public launch, but the benchmark story is still missing

By BurmDesk

Elon Musk says xAI will make Grok 4.5 public Thursday after SpaceX and Tesla beta feedback. The model is framed as Opus-class and lower cost, but xAI has not yet published pricing, benchmarks, API details, or a model card.

Grok 4.5 launch hero with model facts and benchmark caveat
xAIGrok 4.5GrokAI modelscoding agents

Grok 4.5 gets a Thursday public launch date, but the benchmark story is still missing

Elon Musk says Grok 4.5 will become publicly available on Thursday after positive feedback from customers in the beta program.

The short version of Musk’s new claim is clear: Grok 4.5 is coming tomorrow, it is being framed as an “Opus-class” model, and xAI expects it to be faster, more token-efficient, and cheaper than the model class it is being compared against. The longer version is more useful for builders: the launch has more supporting detail than a one-line date tease, but it still does not have the things teams need most — public benchmarks, final pricing, API documentation, context-window limits, or a model card.

That makes Grok 4.5 a serious Thursday watch, not a settled leaderboard event.

What Musk added today

In a July 8 post, Musk wrote that “based on strong positive feedback from customers in our beta test program,” @SpaceXAI would make Grok 4.5 available to the public “tomorrow.” He described the model as “Opus-class,” but “faster, more token-efficient and lower cost.”

That is the first public release-date signal I could verify from Musk for Grok 4.5. It also narrows the positioning. xAI is not framing this as a generic chat refresh. It is positioning Grok 4.5 against the top agentic frontier tier, with cost and token efficiency as part of the pitch.

The catch: no public rate card or benchmark table accompanied the post. “Lower cost” is still a claim, not a published number.

The earlier training details are more specific

The stronger technical details come from a June 28 Musk post that was later quoted and summarized by other X users. In that post, Musk said Grok 4.5 is based on xAI’s 1.5 trillion-parameter V9 foundation model, with Cursor data added during supplemental training. He also said the model was already in private beta at SpaceX and Tesla, and that early evaluations showed performance “close to, perhaps exceeding Opus.”

That gives the launch three concrete anchors:

DetailCurrent status
Foundation modelMusk described it as a 1.5T V9 foundation model
Additional training dataCursor data added during supplemental training
Beta customersSpaceX and Tesla, according to Musk’s June 28 post
Public launch timingThursday, July 9, according to Musk’s July 8 post
Competitive claim“Opus-class,” but no public benchmark table yet

The Cursor detail matters because it points directly at coding-agent workflows. If that data was added through supplemental training rather than base pretraining, the most obvious place to look for gains is not trivia, essay writing, or ordinary chat. It is multi-file coding, IDE-style edits, tool use, diff repair, and agentic loops that resemble real Cursor sessions.

There were also web UI traces

There is another signal beyond Musk’s own posts. On July 6, TestingCatalog posted a screenshot from the Grok web interface showing a Grok 4.5 access prompt in the product UI.

That is not the same thing as a release note, and it should not be treated as one. But as a pre-launch trace, it fits the timing: UI copy appears, then Musk says public access is coming two days later.

Taken together, the public evidence suggests Grok 4.5 was not only a private internal branch. It had already reached customer beta, appeared in web UI copy, and now has a public launch date attached.

xAI’s docs have not caught up yet

As of the current xAI developer docs I checked, Grok 4.5 is not listed on the public models page. The page currently highlights Grok 4.3 for general chat, Grok Build 0.1 for code, the Voice API, and the Imagine API.

The same docs list Grok 4.3 at $1.25 per million input tokens and $2.50 per million output tokens, while Grok Build 0.1 is listed at $1.00 input and $2.00 output per million tokens. Those are useful reference points, but they are not Grok 4.5 pricing. xAI has not published the Grok 4.5 rate card yet.

That gap is important because Musk’s launch framing leans heavily on efficiency. If Grok 4.5 is meant to compete with Opus-class models on coding-agent work, the key comparison is not only raw capability. It is cost per successful task, including output tokens, retries, tool calls, and long-context overhead.

What is still unknown

The unknowns are big enough that teams should not treat Thursday as a migration day. They should treat it as an evaluation day.

The main open questions:

  • Benchmarks: xAI has not published public Grok 4.5 results for SWE-bench, TerminalBench, GPQA, AIME, LMArena, or other common comparison points.
  • Pricing: Musk says lower cost, but xAI has not posted Grok 4.5 API prices.
  • API surface: No public model identifier, availability tier, context-window size, rate limits, or tool-use behavior has appeared in the docs I checked.
  • Model card: There is no public Grok 4.5 model card yet.
  • “Opus-class” definition: Musk’s comparison does not specify which Opus version, which tasks, what latency, or how much test-time compute was used.
  • Cursor effect: Cursor data is a meaningful clue, but xAI has not disclosed the amount, mixture, or exact SFT/RL process.

That does not make the launch unimportant. It just keeps the burden of proof on the release artifacts.

Why builders should care anyway

If the current signals hold, Grok 4.5 could be one of the more relevant launches for coding-agent users this summer. The positioning is unusually direct: a 1.5T V9 base, Cursor-flavored supplemental training, customer beta feedback, and an explicit Opus-class comparison.

The best first test will not be a screenshot or a vibes-based chat demo. It will be a private repo eval with the same failure modes developers actually hit: broken tests, ambiguous refactors, dependency errors, multi-file edits, tool-call loops, and long diffs that need review.

If Grok 4.5 is genuinely faster and more token-efficient at that kind of work, it could matter even without beating every frontier model on every benchmark. A cheaper model that finishes more agentic coding tasks per dollar is more useful than a slower model that wins a narrow academic chart.

But the reverse is also true. If the “Opus-class” claim only holds under internal conditions or handpicked workflows, the public release will expose that quickly.

What to watch Thursday

The launch should be judged by the artifacts xAI publishes, not just by the headline.

The important signals are:

  1. Whether Grok 4.5 appears in the xAI API docs with a stable model ID.
  2. Whether pricing is lower in practice, especially output-token pricing.
  3. Whether xAI publishes a model card with training, safety, and benchmark details.
  4. Whether Cursor-style coding workflows improve, not just single-turn coding prompts.
  5. Whether independent evals reproduce the Opus-class claim within the first few days.

Until those land, Grok 4.5 is best described as a promising, partly specified launch. Musk has given the date and the pitch. The public release still has to supply the proof.

Bottom line

Grok 4.5 has more substance behind it than a simple teaser. Musk has now tied it to a Thursday public launch, a 1.5T V9 foundation model, Cursor supplemental training, SpaceX and Tesla beta testing, and an Opus-class performance claim.

The missing pieces are the pieces that matter most for adoption: public benchmarks, final pricing, model limits, API access details, and independent validation. If xAI publishes those on Thursday, Grok 4.5 could become a real coding-agent contender. If it does not, the launch will start as another frontier-model claim waiting for receipts.

Source notes