Publication date: July 13, 2026
xAI has released Grok 4.5, the company’s most capable model to date, built for coding, agentic tasks, and knowledge work. Grok 4.5 has a 500K-token context window, is served at 80 tokens per second, and is priced at $2 per million input tokens and $6 per million output tokens. xAI positions the model against Opus 4.8, GPT-5.5, and Fable 5, claiming leading results on several software engineering benchmarks at roughly twice the token efficiency of comparable models.
What can Grok 4.5 do?
Grok 4.5 is a text-and-image input model with text output, a 500,000-token context window, function calling, structured outputs, web search, X search, code execution, and configurable reasoning. It was trained on datasets spanning coding, science, engineering, and math, and xAI says it was trained alongside Cursor, which now offers the model on all plans.
The model is aimed at real engineering work rather than benchmark demos. xAI highlights end-to-end app building from a single prompt, including Rust and C/C++ tasks, and reports that Grok 4.5 scores first on Harvey’s Legal Agent Benchmark. Grok 4.5 is also the default model in Grok Build, where it can build multi-sheet Excel models, PowerPoint decks using native shapes, and Word documents.
Grok 4.5 benchmarks and technical specs
On DeepSWE 1.0, measured within each provider’s own harness, Grok 4.5 scored 62.0%, ahead of Opus 4.8 max at 55.75% but behind Fable max at 66.1% and GPT-5.5 xhigh at 64.31%. On the DeepSWE 1.1 run by DataCurve using the mini-swe-agent harness, Grok 4.5 scored 53%, below Opus 4.8 max at 59%. On Terminal Bench 2.1, Grok 4.5 scored 83.3%, close to GPT-5.5 xhigh at 83.4% and Fable max at 84.3%, and above Opus 4.8 max at 78.9%. On SWE Bench Pro, Grok 4.5 reached 64.7%, behind Fable max at 80.4% and Opus 4.8 max at 69.2%.
Grok 4.5 was trained across tens of thousands of NVIDIA GB300 GPUs, with reinforcement learning spanning hundreds of thousands of tasks centered on multi-step software engineering. xAI emphasizes per-token intelligence: the model resolves an average SWE Bench Pro task using 15,954 output tokens, which it frames as about 4.2 times fewer than Opus 4.8 max at 67,020 tokens.
How does Grok 4.5 compare to Opus 4.8 and GPT-5.5?
xAI’s headline claim is efficiency rather than a clean sweep of the benchmarks. Grok 4.5 leads Opus 4.8 on DeepSWE 1.0 and Terminal Bench 2.1 but trails it on DeepSWE 1.1 and SWE Bench Pro. Against GPT-5.5 xhigh and Fable max, Grok 4.5 is competitive on Terminal Bench 2.1 but sits behind on the SWE benchmarks.
The stronger differentiator is cost and speed. At $2 input and $6 output per million tokens, served at 80 tokens per second, and using roughly half the tokens of comparable models on the same tasks, xAI argues Grok 4.5 delivers the highest intelligence per unit of time and cost. Buyers comparing frontier coding models will weigh that efficiency against Opus 4.8’s higher resolve rates on SWE Bench Pro and DeepSWE 1.1.
Grok 4.5 availability and pricing
Grok 4.5 is available now in Grok Build, in Cursor on all plans, and through the SpaceXAI console via the model name grok-4.5. API pricing is $2 per million input tokens, $0.50 per million cached input tokens, and $6 per million output tokens, with higher-context pricing applying above 200,000 tokens. xAI is offering free Grok 4.5 usage for a limited time in Grok Build and Cursor.
One caveat for European users: Grok 4.5 is not yet available in the EU in any SpaceXAI products or the API console. xAI expects EU availability in mid-July 2026.
Conclusion
With Grok 4.5, xAI delivers a coding-focused model with a 500K context window, 80 tokens per second throughput, and $2/$6 pricing, positioned on efficiency against Opus 4.8, GPT-5.5, and Fable 5.
The full announcement is available on the SpaceXAI newsroom. Official announcement: https://x.ai/news/grok-4-5
