Frontier Models

xAI Grok 4.5 Enters Private Beta at SpaceX and Tesla

The model enters restricted testing inside the two companies after training on the V9 foundation model plus Cursor data, with early results positioned against Claude Opus and plans set for monthly new releases.

By Marcus Vance June 28, 2026 4 MIN READ

Inside a large shared engineering workspace that merges the operational environments of an aerospace campus and an electric vehicle development center, multiple teams of anonymous engineers work at long rows of desks equipped with arrays of flat panel monitors displaying complex code repositories and model evaluation dashboards without any visible lettering or branding. Engineers wearing plain dark clothing sit with their backs to the viewer, focused on keyboards and trackpads while reviewing outputs from an advanced AI system in private beta phase. On one side of the room large windows reveal a distant view of a prototype rocket structure on a test stand alongside a sleek electric sedan parked in an adjacent lot, symbolizing the dual corporate settings. Server racks with blinking indicator lights line the rear wall next to whiteboards covered in abstract diagrams and flowcharts. A central table holds open laptops connected to external drives and tablets showing side by side performance comparisons of language models. The space features exposed industrial ceilings with ductwork, neutral gray carpeting, and scattered ergonomic chairs. Additional details include coffee mugs on desks, cables neatly bundled along the floor, potted plants near the windows, and a subtle mix of metallic and composite material samples displayed on shelves. Engineers gesture toward screens while discussing model training runs that incorporate foundation architecture V9 combined with specialized data sources. The overall atmosphere conveys focused technical evaluation in a restricted access environment where early stage testing occurs before wider deployment, with emphasis on iterative development cycles and competitive benchmarking against other frontier systems. Every element remains grounded in real hardware, furniture, and architectural features typical of technology campuses, avoiding any symbolic or abstract representations. The scene captures a single frozen moment of collaborative review involving hardware interfaces, data storage units, and multi monitor workstations arranged to facilitate private internal validation of the Grok 4.5 system within the specified organizations. — Illustration: AI Intel Report

Grok 4.5 is the xAI model now undergoing private beta testing inside SpaceX and Tesla after training on the 1.5 trillion parameter V9 foundation model with added Cursor data.

The announcement of the private beta appeared in a post on X that also outlined training details and future release plans. Reporting from NDTV Profit confirmed the same core facts about the deployment and performance comparisons.

What background context led to the Grok 4.5 private beta?

xAI has pursued rapid model iteration with an emphasis on internal validation at affiliated companies. SpaceX and Tesla supply operational environments that demand high reliability across complex tasks.

The approach allows direct feedback from teams working in aerospace and automotive domains. Prior models in the Grok line established the pattern of using such settings for initial refinement before wider exposure.

What defines the private beta phase in detail?

Grok 4.5 is now in private beta at SpaceX and Tesla. Early evaluations conducted in these settings produced results close to or perhaps exceeding those of Claude Opus.

The restricted access supports focused monitoring and iterative adjustments based on real usage patterns. This phase precedes any broader availability and centers on internal performance validation.

What technical specifics apply to the model and its training?

Grok 4.5 rests on the V9 foundation model that contains 1.5 trillion parameters. Supplemental training incorporated Cursor data to address targeted capability areas.

Reinforcement learning continues to deliver significant gains in model outputs. The Grok Build harness receives ongoing daily enhancements that support overall system stability.

Specification	Description
Foundation Model	V9 with 1.5 trillion parameters
Supplemental Training	Cursor data
Current Status	Private beta at SpaceX and Tesla
Performance Indicator	Close to or exceeding Claude Opus
Ongoing Process	Reinforcement learning and harness improvements

What market and stakeholder implications follow from the beta?

Internal use at SpaceX and Tesla directs the model toward specialized workflows in manufacturing and space operations. Stakeholders gain early access to tools that may improve data processing and decision support in those sectors.

The performance positioning relative to Claude Opus could influence how other frontier model developers set internal targets. Companies may monitor the monthly release cadence for signals about development velocity in the sector.

What expert reactions address the Grok 4.5 announcement?

The primary communication from xAI leadership notes the contributions of the team involved in training and evaluation. It also flags the continued role of reinforcement learning in driving further gains.

Grok 4.5, based on our 1.5T V9 foundation model, with Cursor data added in supplemental training, is now in private beta at SpaceX & Tesla. Early evals show performance close to, perhaps exceeding Opus. RL is continuing to significantly improve the model, and the Grok Build harness gets better every day. Nice work by all those involved! Completely trained from scratch new models will be released by @SpaceX every month this year.Elon Musk, CEO of xAI, Tesla and SpaceX

What comes next in the release schedule?

Completely new models trained from scratch will be released every month this year by SpaceX. The schedule reflects an intent to maintain a steady pace of foundation-level updates.

Ongoing reinforcement learning is expected to produce incremental improvements between releases. Internal testing at the two companies will likely inform adjustments ahead of each new model.

Continue private beta evaluation at SpaceX and Tesla
Apply reinforcement learning for further performance gains
Release new scratch-trained models monthly via SpaceX
Refine the Grok Build harness on a daily basis

Frequently asked

Where is Grok 4.5 currently being tested?

Grok 4.5 is in private beta at SpaceX and Tesla. The testing remains internal to these organizations.

What performance level has Grok 4.5 reached relative to other models?

Early evaluations show performance close to or perhaps exceeding Claude Opus. Reinforcement learning is continuing to improve the model.