xAI has pushed Grok 4.5 into private beta at both SpaceX and Tesla, running on a 1.5-trillion-parameter V9 foundation model that represents a threefold jump from the 500-billion-parameter v8-small currently powering Grok on X. Elon Musk announced the beta on June 28 via X, describing early internal evaluations as showing performance “close to, perhaps exceeding Opus.”
The model is not available to the public. No third-party benchmarks exist. The only data points come from xAI’s own teams at SpaceX and Tesla, plus one anecdotal developer report from Mehul Mohan who tested an early build and described it as “similar to Opus.”
What Makes Grok 4.5 Different From Previous Versions
The V9 foundation model that underlies Grok 4.5 completed training on May 26, 2026. It is a ground-up redesign of xAI’s architecture, distinct from the v8-small that handles production Grok traffic on X.
Three technical details stand out:
Scale: At 1.5 trillion parameters, Grok 4.5 is approximately three times larger than the production model and 50% larger than Grok 4.4 (roughly 1 trillion parameters, shipped in late May 2026). That is a significant jump in one month.
Cursor training data: SpaceX acquired Anysphere (Cursor’s parent company) for $60 billion in June 2026 and has been feeding Cursor IDE session data into Grok training. For Grok 4.5, Cursor coding data was added in supplemental post-pre-training to sharpen code generation and technical reasoning.
Deployment strategy: xAI is using SpaceX’s aerospace engineering workflows and Tesla’s vehicle software development as real-world evaluation environments before any public release. These are harder tasks than standard benchmarks.
Monthly Model Releases Through End of 2026
Musk confirmed that SpaceX plans to release “completely new models trained from scratch” every month through the end of 2026. That would make xAI the fastest-iterating frontier lab by cadence, with Grok 4.6, 4.7, and beyond shipping monthly as stepping stones toward Grok 5.
Grok 5 is targeting 6 to 10 trillion parameters, which would make it the largest model architecture ever publicly discussed. It is currently training on Colossus 2 alongside six other concurrent training runs.
The Grok Build coding harness, xAI’s internal tool for evaluating model performance on real engineering tasks, runs daily improvement cycles on Grok 4.5 during the beta period. This gives xAI continuous feedback loops that most labs cannot replicate because they lack integrated production environments.
The Competitive Landscape Shifts
The monthly iteration cycle puts pressure on every other frontier lab. Anthropic just shipped Claude Sonnet 5 and is preparing for its October 2026 IPO. OpenAI is planning a September 2026 IPO while reportedly offering the US government a 5% stake. Google launched Gemini 3.5 at I/O 2026.
None of these companies has committed to monthly model releases. Whether xAI’s pace produces compounding capability gains or hits diminishing returns is the empirical question the second half of 2026 will answer.
The independent benchmark gap matters here. Without public evaluation data, Grok 4.5’s “close to Opus” claim remains xAI’s internal assessment. Opus currently leads the Artificial Analysis Intelligence Index at 61.4 and scored 69.2% on agentic coding benchmarks.
What Happens Next
xAI has not announced a public release date for Grok 4.5. The SpaceX and Tesla beta is internal-only. The next model in the pipeline, already in training, targets 2 trillion parameters and will incorporate Cursor data from the initial pre-training stage rather than as a supplement. xAI expects this to produce stronger coding performance than the current model.
The monthly cadence means Grok 4.6 could arrive as early as late July 2026. Whether any of these models reach public availability before year-end depends on how the SpaceX and Tesla evaluations progress.
Frequently Asked Questions
Is Grok 4.5 available to the public?
No. As of July 2026, Grok 4.5 is in private beta restricted to SpaceX and Tesla employees. There is no public API, no access through X (formerly Twitter), and no announced release date for general availability.
How does Grok 4.5 compare to Claude Opus?
The only performance claims come from xAI’s internal evaluations, which Elon Musk described as “close to, perhaps exceeding Opus.” There are no independent benchmarks. Claude Opus currently leads the Artificial Analysis Intelligence Index at 61.4. Until third-party data emerges, the comparison cannot be verified.
What is Cursor’s role in Grok 4.5 training?
SpaceX acquired Cursor’s parent company Anysphere for $60 billion in June 2026. Cursor IDE session data was added to Grok 4.5’s supplemental training to improve coding and technical reasoning performance. The next model will incorporate Cursor data from the initial pre-training stage.
When will Grok 5 be released?
Grok 5 is targeting 6 to 10 trillion parameters and is training on Colossus 2. No release date has been announced. The monthly V9 variants (4.5, 4.6, 4.7) through Q4 2026 are stepping stones in the capability improvement process that feeds Grok 5.
Can I run Grok 4.5 locally?
No. At 1.5 trillion parameters, Grok 4.5 requires massive compute infrastructure to run. It is not open-weight and there are no plans to release the weights. This is different from open-weight models like LongCat-2.0 or DeepSeek V4-Pro, which are available for self-hosting.
