Grok 4.5 Arrives: xAI's Cursor-Trained Model Challenges the Cost-Performance Curve
News Summary
xAI has officially launched Grok 4.5, its most capable model to date, marking a significant step forward in the company's push to compete at the frontier of large language model development. The release, announced on July 8, 2026 (Pacific Time), positions Grok 4.5 as an "Opus-class" system that xAI founder Elon Musk describes as faster and more token-efficient than comparable frontier models, while being offered at a notably lower price point.
Availability and Access
Grok 4.5 is available starting today through Grok Build, across all plans in the Cursor coding editor, and via the xAI developer console. The model is rolling out globally, though it is not yet accessible in the European Union through xAI products or the API console; EU availability is expected by mid-July 2026 (Central European Time), pending regional review.
Technical Foundation
The new model is built on a 1.5-trillion-parameter foundation internally referred to as V9. Training was carried out across tens of thousands of NVIDIA GB300 GPUs, with xAI engineers developing new stability and optimization techniques to support runs at this scale. Notably, Grok 4.5 was trained in close collaboration with Cursor, the AI-assisted coding platform that xAI agreed to acquire in a deal valued at $60 billion in June 2026. Real developer session data from Cursor — including debugging traces, multi-file code diffs, and user correction patterns — was incorporated into the training pipeline, giving the model practical grounding in real-world software engineering workflows.
Capabilities and Demonstrations
xAI highlighted the model's coding and agentic reasoning abilities with demonstrations such as generating a full interactive Three.js solar-system simulation, complete with adjustable time controls and a styled heads-up display, from a single natural-language instruction. The company says Grok 4.5 was trained on a broad mix of data spanning coding, science, engineering, and mathematics, enabling it to handle complex, multi-step technical tasks with greater coherence.
Benchmark Performance
According to benchmarks published by xAI, Grok 4.5 outperforms comparable frontier models on two of four disclosed evaluations — DeepSWE 1.0 and Terminal-Bench 2.1 — while trailing on DeepSWE 1.1 and SWE-Bench Pro by modest margins. A standout efficiency result came from SWE-Bench Pro, where Grok 4.5 reportedly resolved tasks using an average of roughly 15,954 output tokens, compared to about 67,020 tokens used by a leading rival model on the same benchmark, suggesting a substantial efficiency advantage in reasoning-heavy coding tasks. The model also ranked first on Harvey's Legal Agent Benchmark, a notable result for a model marketed primarily around software engineering, suggesting its training approach generalizes well beyond coding tasks alone.
Pricing and Performance
xAI has priced Grok 4.5 at $2 per million input tokens and $6 per million output tokens, undercutting several competing frontier models. The model operates with a configurable reasoning effort setting — low, medium, or high, with high set as the default — and generates output at approximately 80 tokens per second. Grok 4.5 has also become the default model inside Grok Build, xAI's application-building environment.
Industry Context
The release comes amid intensifying competition among frontier AI labs to deliver models that balance raw capability with cost efficiency, particularly for developers and enterprises running large-scale agentic and coding workloads. By tightly integrating training data from a widely used coding tool, xAI is positioning Grok 4.5 as a practical, workflow-aware option for technical teams, rather than a purely benchmark-driven release.