Breaking

Grok 4.5 Begins Internal Beta at SpaceX and Tesla on 1.5T Base Model

June 28, 2026 at 07:47 EDT

Foundation Models
Software Dev & Coding
Robotics

xAI's new model "Grok 4.5" has entered private beta inside SpaceX and Tesla. Built on the 1.5-trillion-parameter "V9" foundation model, early evaluations reportedly show it matching, or possibly exceeding, Anthropic's top-tier Opus model.

June 28, 2026 · xAI

Grok 4.5 goes live in private beta — built on the 1.5-trillion-parameter "V9" model

Running internally at SpaceX and Tesla, early evaluations describe it as close to — and perhaps exceeding — Anthropic's Claude Opus. No public release or API access yet.

1.5T

parameters in the new V9 foundation model

~3×

the scale of the existing Grok 4.2

Monthly

from-scratch models planned this year

Scaling up the foundation model

Parameter count, roadmap by version

~0.5T

Grok 4.2

~1.0T

Grok 4.4

1.5T

Grok 4.5 (V9)

The Cursor data feedback loop

Developers work in Cursor

→

Workflow logs feed training

→

RL sharpens coding model daily

→

Better model loops back to users

Where it stands today

"Close to, perhaps exceeding, Claude Opus" — but no public benchmark scores yet.

Access

Private beta inside SpaceX & Tesla — no public release or API

Training

Multiple models running in parallel on the Colossus GPU cluster

Unverified

SWE-bench Verified scores not yet disclosed; top models sit in the high-80% range

Optimistic

Combining the 1.5T V9 base with Cursor data is expected to lift both reasoning and coding, with RL fine-tuning sharpening the model each iteration.

Skeptical

The aggressive monthly from-scratch cadence has drawn half-joking doubt, and hands-on accounts from real-world use remain limited.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.