Breaking

SpaceXAI Rolls Out Grok 4.5 for Coding and Agentic Work

July 8, 2026 at 14:30 EDT

Foundation Models
Software Dev & Coding
AI Agents

SpaceXAI (formerly xAI) on July 8, 2026 announced a new model, Grok 4.5, built for coding, agentic tasks and knowledge work, via its official news page. The company stresses that it delivers "the highest intelligence per unit of time and cost."

Jul 8, 2026 · SpaceXAI

Grok 4.5: Opus-class power, built for speed and cost

A new model for coding, agentic tasks and knowledge work — claiming "the highest intelligence per unit of time and cost," now live across Grok Build, Cursor, the xAI API and console.

~1.5T

V9 foundation model parameters

80 tok/s

Up to, generation speed

4.2×

Fewer tokens than Opus 4.8 Max on SWE-Bench Pro (vendor claim)

Token efficiency vs Opus 4.8 Max

Tokens used on SWE-Bench Pro — shorter is better (vendor claim)

4.2×

Opus 4.8 Max

1×

Grok 4.5

Input

per 1M tokens

Output

per 1M tokens

Available on

Grok Build · Cursor · xAI API · console (API excludes EU)

The data flywheel

Real engineering work
SpaceX · Tesla · Cursor

→

Coding data
supplemental training

→

Better model
Grok 4.5

✓ What developers praise

Opus-level capability at high speed and low cost; practical for long-running agentic coding and knowledge work. SpaceX and Tesla engineers found it "genuinely useful."

! The open question

Independent benchmarks (LMArena, SWE-bench, Artificial Analysis) remain unpublished — claims are vendor-stated. Some worry raw reasoning may trail Fable 5 or Claude Opus 4.8.

Looking ahead

SpaceXAI plans to ship new "from-scratch" models on a monthly cadence through late 2026 — putting its frontier iteration speed, and third-party verification of its claims, in the spotlight.

Continue reading

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