SpaceXAI released Grok 4.5, its coding-focused flagship model now available in Grok Build, Cursor, and the company's API console.

The model was trained specifically for coding and agentic workflows in partnership with Cursor, marking SpaceXAI's first model built beyond general chat applications. EU access remains unavailable, with rollout expected mid-July.

Grok 4.5 serves as the default model in Grok Build, where it operates as a coding agent through CLI, terminal UI, scripts, and Agent Client Protocol integrations. The model also runs inside Microsoft Office add-ins for Word, PowerPoint, and Excel, creating spreadsheet models and document drafts.

The model supports text and image input with a 500,000-token context window, function calling, structured outputs, web search, X search, and code execution. Developers can configure reasoning effort across low, medium, or high settings, with high reasoning enabled by default.

Pricing and performance benchmarks

API pricing sits at $2 per million input tokens, $0.50 per million cached tokens, and $6 per million output tokens. Higher-context pricing applies above 200,000 tokens.

SpaceXAI published benchmark results showing Grok 4.5 scored 62.0% on DeepSWE 1.0, 53% on DeepSWE 1.1, 83.3% on Terminal Bench 2.1, and 64.7% on SWE Bench Pro. The company claims 80 tokens per second serving speed and uses 15,954 output tokens on average for SWE Bench Pro tasks.

The model was trained across tens of thousands of NVIDIA GB300 GPUs using data filtering, deduplication, and reinforcement learning over hundreds of thousands of tasks. SpaceXAI targeted multi-step software engineering work through automated grading and long-running agentic rollouts.

Grok 4.5 ranks fourth on GDPval-AA v2 with an Elo of 1543, trailing only recent Claude releases from Anthropic on real-world agentic knowledge work tasks.