Term
Grok 4.5
Grok 4.5 is xAI's July 2026 frontier model for coding, agentic tasks and knowledge work — high token efficiency at $2 input and $6 output per 1M tokens.
Grok 4.5 — explained in more detail
xAI introduced Grok 4.5 in July 2026 as its strongest model at the time. It is designed for coding, agentic tasks and knowledge work and succeeds version 4 within the Grok line. xAI positions the model on efficiency: on SWE-Bench Pro tasks it uses, per the vendor, roughly 4.2 times fewer output tokens than Opus 4.8 (max) and solves tasks in fewer steps. Reasoning depth is configurable (low, medium, high; default high), letting cost and thoroughness be tuned per use case.
In coding benchmarks xAI reports, among others, a 64.7% resolve rate on SWE-Bench Pro and 62.0% on DeepSWE 1.0. Throughput is around 80 tokens per second.
Example / Practical use
The key data: pricing is $2 per 1M input tokens and $6 per 1M output tokens; larger prompts (above 200,000 tokens) move to higher tiered rates. Access is proprietary via the xAI API console, as the default model in Grok Build, and in Cursor across all plans; for end users, Grok 4.5 is included in the SuperGrok subscription. The high token efficiency targets coding agents that run many steps — there, less output per solved task noticeably lowers total cost. A context window of around 500,000 tokens is also cited.
Distinction
Grok is xAI’s own model family and competes with Claude Opus, Gemini Pro and GPT. Version 4.5 is the evolution of Grok 4 with a focus on efficiency and agent suitability rather than raw capability alone. The explicit reference point is the Opus class: xAI markets Grok 4.5 not as the single most capable model on the market but as the cheaper one per solved task.
Discover more
GPT-6 Astra Found Questions My First Security Review Missed
I used GPT-6 Astra and Fable 5.1 as independent reviewers for RLS, API and tenant-isolation checks. The useful part was the structured cross-review.
GlossaryGrok 4.6
Grok 4.6 (August 2026) is xAIs flagship model for long-running agentic work — 500k-token context, text and image input, tiered pricing from 2/6 USD per million tokens.
EncyclopediaLLM Model Families at a Glance
Claude, GPT, Gemini, Llama & co. — who builds what, where each family shines, and how to pick the right model for your own use case.