xAI Grok 4.5
A coding and agentic flagship trained alongside Cursor, priced aggressively at $2 per million input tokens and $6 per million output — more than 60% cheaper than Opus 4.8 or GPT-5.5 — with a 500K-token context window (down from Grok 4.3's 1M) and configurable reasoning effort. Independent scores place it 4th on the Artificial Analysis Intelligence Index (54), with roughly 2x the token efficiency of competing flagships on agent benchmarks. Note the vendor now brands itself SpaceXAI.
Public launch July 8 after a private beta, available in Grok Build, Cursor on all plans, the API console, and OpenRouter. Not available in the EU at launch. The Grok Build environment was open-sourced under Apache 2.0 on July 16.
Reality review pending — we publish the executive verdict after the launch has had time to meet the real world.
A direct price attack on the coding-agent market: near-frontier quality at roughly a third of incumbent prices, with token efficiency that compounds the savings on agent workloads. Enterprises negotiating coding-AI spend now have a credible lever — the caveats are the smaller context window and EU unavailability at launch.