PALO ALTO, Calif.: xAI releases Grok 4.5 to the public on July 8, pricing it to undercut every frontier rival.
Input costs $2 per million tokens, output $6. The context window runs 500,000 tokens, enough for long coding sessions and large repositories. The launch beats OpenAI’s GPT-5.6 general availability by exactly one day. Nobody believes that timing is coincidental.
Under the hood sits a 1.5 trillion-parameter Mixture-of-Experts model. The interesting detail is the training data: xAI trained partly on interaction data from Cursor, the AI coding editor. That gives the model an unusual amount of exposure to how real developers actually work.
The pitch is efficiency rather than raw capability. Grok 4.5 scores 83.3% on Terminal-Bench 2.1 for agentic terminal tasks. It does so using roughly a quarter of the output tokens Claude Opus 4.8 burns on similar work. For teams running high-volume agentic pipelines, token thrift is money.
Artificial Analysis ranks it fourth overall with an Intelligence Index of 54. Fourth place at a fraction of the price is a viable business. Independent benchmark verification is still pending.
The release continues xAI’s rapid cadence under the SpaceX umbrella. The company folded into SpaceX before June’s record $75 billion IPO. The S-1 disclosed xAI lost $2.4 billion in the first quarter alone, so cheap pricing carries obvious strategic risk. Consumer distribution still runs through X, where Grok serves millions daily.
The launch landed in the densest release week the industry has seen. Meta’s Muse Spark 1.1 and the full GPT-5.6 family arrived within 24 hours. Anthropic’s Claude Fable 5 had returned to availability the week before.
Previous
