What Is Grok 4.5?
Grok 4.5 is xAI’s flagship model, released publicly on July 8–9, 2026 — xAI now operates as SpaceXAI following SpaceX’s earlier merger with the company. Built on a new 1.5-trillion-parameter “V9” foundation, it’s the first Grok model co-trained using real Cursor developer session data, reflecting SpaceX’s parallel $60 billion acquisition of Cursor. Elon Musk pitched it as “an Opus-class model, but faster, more token-efficient and lower cost” — and while xAI’s own benchmark chart tells a more mixed story than that framing suggests, the price-performance story is genuinely strong.
Key Features
- 500K-token context window — smaller than GPT-5.6 Sol’s ~1.05M or Claude’s context, but ample for most agentic and coding workloads.
- Configurable reasoning effort — a reasoning dial you can set per call, trading latency for depth on harder tasks.
- Grok Build — Grok 4.5 is the default model here, handling end-to-end app builds, multi-sheet Excel workbooks with live web research, and native Word/PowerPoint generation from a single prompt.
- Native web and X search — built-in real-time data access, a long-standing Grok differentiator.
- Cursor integration — live inside Cursor on all plans, reflecting the co-training relationship between the two companies.
Benchmarks
On the independent Artificial Analysis Intelligence Index, Grok 4.5 scores 54 and ranks #4 overall — behind Claude Fable 5 (60), Claude Opus 4.8 (56), and GPT-5.5 (55), but ahead of every open-weight model and all current Gemini models. It takes the single top spot on agentic tool use specifically. On coding benchmarks: 83.3% on Terminal-Bench 2.1, 64.7% on SWE-Bench Pro, and 78.0% on SWE-Bench Multilingual (per Cursor’s own launch data). It also ranked #1 on Harvey’s Legal Agent Benchmark, an independent evaluation of 1,200+ practical legal tasks.
Pricing
Grok 4.5 API pricing: $2.00 per million input tokens, $6.00 per million output tokens, with cached input at $0.50 per million (a 75% discount). That undercuts Claude Opus 4.8 ($5/$25) by roughly 3x on input and 4x on output, and runs well below GPT-5.6 Sol’s $5/$30. Consumer access comes through Grok’s SuperGrok tiers rather than a separate API-only plan. One limitation worth flagging: Grok 4.5 was not available in the EU at launch, with EU access expected around mid-July 2026 — confirm current availability before building for European users.
How It Compares
Against GPT-5.6 Sol ($5/$30), Grok 4.5 is dramatically cheaper but scores lower on the Intelligence Index and ships a smaller context window. Against Claude Opus 4.8, the “Opus-class” marketing framing is best read as “near-frontier at a fraction of the price” rather than a like-for-like capability match — it trails on raw benchmarks and carries a real hallucination-rate caveat independent reviewers have flagged. Against OpenCode or Cursor as coding environments, Grok 4.5 is a model you’d route through those tools rather than a competing IDE — and its native Cursor integration makes that pairing especially easy.
Is Grok 4.5 Worth It?
For cost-sensitive, high-volume agentic and coding workloads, Grok 4.5’s price-to-performance ratio is hard to beat among frontier-tier options — you’re getting roughly GPT-5.5-adjacent quality on agentic tool use at a fraction of the token cost. It’s not the model to reach for when you need the absolute ceiling on reasoning quality; Claude Opus 4.8 and Fable 5 still lead there. The practical move: run your own eval set — your real prompts, your real code — against Grok 4.5 and whatever you’re currently using, and compare quality against total token cost per task before committing production workloads.
Last updated: July 17, 2026.
Related reading
- SpaceX’s $60B Cursor acquisition — the deal behind Grok 4.5’s Cursor co-training
- GPT-5.6 GA breakdown — the pricier frontier alternative