Models · 8 Jul 2026 · 2 min read

SpaceXAI Releases Grok 4.5 to Challenge Opus on Price

The new model claims Opus-class performance at a fraction of the cost, forcing builders to re-evaluate their default API choices for workhorse tasks.

Pen-and-ink illustration: a small, perfectly cut gemstone resting. For the story "SpaceXAI Releases Grok 4.5 to Challenge Opus on Price".
— Pen-and-ink illustration: a small, perfectly cut gemstone resting. For the story "SpaceXAI Releases Grok 4.5 to Challenge Opus on Price". —

What happened

SpaceXAI released its latest model, Grok 4.5, on Wednesday. In a blog post, the company positioned it as a workhorse model for coding, research, and writing. According to TechCrunch AI, founder Elon Musk described the model as 'Opus-class' but faster and more token-efficient.

SpaceXAI priced Grok 4.5 at $2 per million input tokens and $6 per million output tokens. This significantly undercuts competitors like Anthropic's Opus 4.7 and OpenAI's most capable models.

How the room's reading it

The conversation on X, led by Elon Musk's own posts, frames Grok 4.5 as a direct challenger to Anthropic's Opus on both performance and cost. Developers are focusing on the pricing — $2 per million input tokens is aggressive, especially when paired with the claim of double the token efficiency. This puts immediate pressure on both Anthropic and OpenAI.

Some observers are pointing out the timing, with the release landing just a day before OpenAI's planned drop of GPT 5.6. The consensus among practitioners seems to be cautious optimism, waiting to see if real-world performance matches the marketing claims and benchmark sheets.

Sailfish's take

We're not looking at the benchmarks. The price point is the real headline here — it's designed to make teams re-evaluate their default API calls for routine tasks. Calling a model 'Opus-class' is easy, but delivering that quality at this cost is another matter entirely. We've shipped enough products to know that token efficiency claims often don't survive contact with real-world use cases.

We'd suggest spinning up a parallel evaluation process for Grok 4.5 on a non-critical workload. It's not worth switching production traffic yet, but it's absolutely worth spending a few dollars to see if the cost-per-decent-answer holds up.

Our take — your read?

Be the first to weigh in.

Sources
— END OF DISPATCH — Models