Grok 4.5

https://x.ai/news/grok-4-5

$2 per M input tokens, $6 per M output tokens via xAI API

$2/$6 per million tokens — undercuts Opus 4.7 and GPT-5.6 Sol pricing
64.7% SWE-bench Pro resolve rate
83.3% Terminal Bench 2.1 (tied with GPT-5.5)
29.0% SWE Marathon resolution rate — best in class
4.2x token efficiency vs Opus 4.8 (15,954 vs 67,020 avg output tokens)
500K context window
80 TPS inference speed
Available in Grok Build CLI (default) and Cursor (all plans)
Trained on coding, science, engineering, and math data with Cursor collaboration
Not yet available in EU

Token-efficiency advantage: 4.2x fewer output tokens than Opus 4.8 per task
Aggressive pricing undercuts Anthropic and OpenAI flagship models
Strong Terminal Bench 2.1 score (83.3%) — tied with GPT-5.5
Best-in-class SWE Marathon resolution rate (29.0%)
Default model in Grok Build CLI and available in Cursor on all plans
80 TPS inference speed — fast token delivery
500K context window handles substantial codebases

SWE-bench Pro 64.7% trails Fable (80.4%) and Opus 4.8 (69.2%)
"Opus-class" marketing claim is optimistic on raw benchmark performance
Not yet available in the EU (expected mid-July 2026)
Context window (500K) trails Claude Opus and GPT-5.4 (1M+)
Released same week as GPT-5.6 — faces stiff competition