xAI Releases Grok 4.5, a Coding and Agent Model Trained With Cursor at Half Rivals' Price

xAI has released Grok 4.5, which the company says is the first model in its lineup trained from the start for software engineering and autonomous agents rather than general conversation. It shipped simultaneously inside Cursor, the code editor whose real developer sessions xAI says it used as additional training data. That means the model was exposed to actual debugging, refactoring, and test-writing traces rather than synthetic code.

The commercial signal is the price. Grok 4.5 lists at two dollars per million input tokens and six dollars per million output tokens, with a 500,000-token context window and adjustable reasoning depth. That is far below what the leading closed models charge for output tokens. xAI's own figures claim performance on par with GPT-5.5 and approaching Anthropic's Opus 4.8 on coding, plus a first-place ranking on Harvey's Legal Agent Benchmark.

Those claims come from a vendor announcement, not from independent reproduction. GPT-5.5 and Opus 4.8 are the current closed-model leaders, and matching them at a third of the output price would be a real gain in capability per dollar rather than a single benchmark number. Independent tests on public agentic-coding suites, together with the questions about evaluation integrity raised elsewhere in today's edition, will determine whether the parity claim holds when third parties test it.