Grok 4.5 nearly matches GPT-5.5 on coding benchmarks — and costs a fraction of the price

Elon Musk's AI company xAI has released Grok 4.5, a model built specifically for coding and engineering work. It scores 83.3% on Terminal Bench 2.1 — within a hair of GPT-5.5 (83.4%) and just behind Fable 5 (84.3%). The bigger story is price: Grok 4.5 completes a typical coding task for $2.49, compared to $5.07 for GPT-5.5 and $11.80 for Fable 5 running on Claude Code.

What changed

Grok 4.5 is trained specifically on coding, engineering, and math tasks, using tens of thousands of NVIDIA GB300 GPUs. The previous version, Grok 4.20, scored zero on agentic terminal benchmarks — tests where an AI must plan multi-step tasks inside a command-line environment, not just predict the next word. That jump from nothing to near-the-top in one generation is the core of xAI's official launch claim.

Token efficiency backs the cost argument: Grok 4.5 uses 4.2 times fewer output tokens than Anthropic's Opus 4.8 on comparable coding tasks (15,954 average vs. 67,020). For teams running hundreds of automated engineering jobs a day, that gap compounds fast.

The competitive picture

On the Terminal Bench leaderboard, the gaps are narrow — Grok 4.5 trails Fable 5 by just one percentage point. Where Grok lags more clearly is on DeepSWE 1.1, where GPT-5.5 scores 67% against Grok's 53%. Benchmark rankings at this tier shift constantly, so the capability comparison is less stable than the pricing one.

An independent evaluation from Snorkel GDPVal+ paints a stronger picture for professional work: Grok 4.5 hits 29% on complex workplace reasoning tasks versus 22% for GPT-5.5 and 21% for Opus 4.8. The widest gaps appear in legal and healthcare queries — 40% vs. 27–28% and 35% vs. 23–25%, respectively.

One reliability caveat worth noting: Grok 4.5's hallucination rate rose to 54% on the AA-Omniscience Index, compared to 52% for rivals. That's a real concern for production use where accuracy matters more than throughput.

Where it's available

Grok 4.5 is available now in Grok Build (xAI's coding CLI), Cursor, and the xAI API console, priced at $2 per million input tokens and $6 per million output tokens. xAI acquired Cursor in June 2026 for $60 billion in equity, giving it direct access to real developer session data — and making Cursor the primary integration target. Users in the EU won't have access until mid-July 2026, as xAI works through compliance requirements.