xAI · 2026-07-08 · seismic
Grok 4.5 — xAI's Opus-class flagship at $2/$6 per million tokens
Grok 4.5 lands on the xAI API, Cursor, and Grok Build at $2 input / $6 output per 1M tokens. Trained with Cursor on tens of thousands of GB300 GPUs, it hits 83.3% on Terminal-Bench 2.1.
xAI's new coding flagship ties GPT-5.5 on Terminal-Bench at a quarter of the price of Fable 5 and Opus.
Key specs
| Swe bench pro | 64.7% |
|---|---|
| Terminal bench 2.1 | 83.3% |
Quick facts
| Maker | xAI (SpaceXAI) |
|---|---|
| Model ID | grok-4.5 |
| Context window | 500K tokens |
| Price (input) | $2 / 1M tokens |
| Price (output) | $6 / 1M tokens |
| Reasoning effort | Low, medium, or high (default high) |
| Availability | xAI API, Cursor, Grok Build, iOS, CLI, SDK (EU pending) |
Benchmarks
Pricing
| Input | $2.00 / 1M tokens |
|---|---|
| Output | $6.00 / 1M tokens |
| Fast input · grok-4.5-fast variant | $4.00 / 1M tokens |
| Fast output · grok-4.5-fast variant | $18.00 / 1M tokens |
What is it?
Grok 4.5 is xAI's new flagship coding and agent model, released July 8, 2026. Built on the 1.5 trillion parameter V9 foundation with a mixture-of-experts design, the model handles a 500K token context and streams around 80 tokens per second. It ships simultaneously on the xAI API, Cursor, Grok Build, and mobile.
How does it work?
Grok 4.5 was trained jointly with Cursor on trillions of tokens of coding data, then reinforcement-learned inside distributed agent environments that Cursor and xAI built to mimic real long-running engineering tasks. A configurable reasoning_effort parameter (low, medium, or high, default high) lets callers dial thinking time per request. Function calling, web and X search, and code execution are supported out of the box.
Why does it matter?
Aggressive pricing on Grok 4.5 collapses the cost gap between frontier coding models: at $2 in / $6 out per million tokens, it undercuts Fable 5 (5x) and Opus 4.8 (roughly 4x) while landing within a point of both on Terminal-Bench. Combined with roughly 4.2x better output-token efficiency versus Opus 4.8 on SWE-Bench Pro, xAI is betting cost per solved task, not headline accuracy, is the metric that now moves developers.
Who is it for?
Cursor users, coding-agent builders, and API teams shopping for cheaper Opus-class inference.
Frequently asked questions
- How much does Grok 4.5 cost?
- Grok 4.5 costs $2 per 1M input tokens and $6 per 1M output tokens on the xAI API. A faster variant runs $4 in / $18 out. That is well below Fable 5 ($10 / $50), Opus 4.8 ($5 / $25), and GPT-5.5 or 5.6 Sol ($5 / $30), which xAI positions as its main competitive lever alongside token efficiency.
- How does Grok 4.5 compare to Claude Opus and GPT-5.5 on coding?
- Grok 4.5 scores 83.3% on Terminal-Bench 2.1, effectively tying GPT-5.5 at 83.4% and trailing Anthropic's Fable 5 at 84.3%. On SWE-Bench Pro it hits 64.7%, ahead of Opus 4.7 (64.3%) but well behind Fable 5 (80.4%). It resolves SWE-Bench Pro tasks with about 4.2x fewer output tokens than Opus 4.8.
- Where can I use Grok 4.5 today?
- Grok 4.5 is live on the xAI API, in Cursor, and inside Grok Build on desktop, web, iOS, CLI, and SDK. SpaceXAI has doubled subscription usage allocations for the first week. The model is not yet available in the EU across xAI products or the API console, with EU availability expected in mid-July 2026.
- How was Grok 4.5 trained?
- Grok 4.5 is a mixture-of-experts model built on xAI's 1.5 trillion parameter V9 foundation, then supplementally trained on trillions of tokens from Cursor plus STEM, engineering, and research data. Training ran on tens of thousands of NVIDIA GB300 GPUs and used reinforcement learning against realistic long-running agent environments Cursor and xAI built together.
Try it
curl https://api.x.ai/v1/chat/completions -H 'Authorization: Bearer $XAI_API_KEY' -d '{"model":"grok-4.5","messages":[{"role":"user","content":"Hello"}]}'