Musk says Grok 4.5 drops tomorrow with Opus-class power

For months, developers have faced the exact same trade-off: pay up for sluggish frontier models, or settle for fast, cheap models that stumble on complex code.

xAI—now operating under SpaceXAI following its public debut a few weeks back—is trying to break that dynamic. Elon Musk announced that Grok 4.5 is rolling out to the public, calling it an Opus-class model that runs faster, costs less, and uses far fewer tokens.

Built for workhorses, trained alongside Cursor

Rather than fixating strictly on top-of-the-chart benchmark wins, SpaceXAI built Grok 4.5 for production work: software engineering, autonomous agents, research, and routine clerical tasks.

Benchmark figures released by the company show performance hovering just shy of best-in-class, but real-world execution is the priority here. The model was trained alongside popular AI coding environment Cursor and delivers inference speeds of 80 tokens per second.

Musk added that SpaceXAI's internal evaluations peg Grok 4.5 as roughly comparable to Anthropic’s Opus 4.7, but significantly faster. The company also claims the model achieves twice the token efficiency of competing flagship offerings, averaging 15,954 output tokens per SWE Bench Pro task.

Grok 4.5 is accessible immediately via Grok Build, Cursor, and the API, with European Union access expected in mid-July.

Slashing API bills to a fraction of rivals

The pricing structure is where SpaceXAI is really putting pressure on competitors.

Grok 4.5 costs $2 per million input tokens and $6 per million output tokens. Compare that to Anthropic’s Opus 4.7, which sits at $5 per million input and $25 per million output tokens. OpenAI’s premier model, GPT-5.6 Sol, runs $5 per million input and $30 per million output.

At $6 for a million output tokens, Grok 4.5 is effectively matching the output price of OpenAI's budget-tier Luna model while claiming top-tier Opus-level intelligence.

Why it matters

If you build agents or write code for a living, this release is worth testing immediately. Massive multi-step workflows often burn through tens of thousands of tokens, turning high-end models into a luxury. Getting near-Opus reasoning at a quarter of the output cost—directly inside Cursor—changes unit economics for startups and dev teams.

For the AI landscape, it signals that the frontier battle is moving from raw benchmark vanity metrics to pure cost-performance and inference efficiency. With OpenAI rolling out GPT-5.6 Sol this same week, SpaceXAI is using its post-IPO momentum to force an aggressive pricing war.

We think pairing directly with Cursor for training was a masterstroke, locking down distribution where developers already live. The risk, of course, is that SpaceXAI’s internal benchmarks admit the model is still just short of best-in-class. Whether that token efficiency holds up without cutting corners on nuanced edge cases will be what decides its enterprise staying power.