Grok 4.5 Ships as xAI's Default Chat Model
xAI launched Grok 4.5 on July 8 — 83.3% on Terminal-Bench 2.1, 4.2× token efficiency, and $2/$6 pricing. helloai's tracked Grok entry moves from 4.3 to 4.5.
On July 8, xAI shipped Grok 4.5 — and helloai's weekly drift check missed it. The updater was watching memory for a Grok 4.4 release that never came, while xAI's docs and news page already listed 4.5 as the recommended chat and code model. Grok 4.5 replaces Grok 4.3 in models.json today: $2/$6 per million tokens, 500K context, and xAI's explicit guidance to use it for everything except dedicated voice, image, and video APIs.
The capability story is engineering speed, not arena rank. xAI claims 83.3% on Terminal-Bench 2.1, 64.7% on SWE Bench Pro, and 29.0% resolution on SWE Marathon — competitive with GPT-5.6 Sol and Claude Opus 4.8 on agentic coding slices. The sharper claim is token efficiency: Grok 4.5 averages 15,954 output tokens per SWE Bench Pro task versus 67,020 for Opus 4.8 (max), a 4.2× gap that compounds in production agent loops. xAI serves the model at 80 tokens per second, positioning it as a fast-model-speed flagship rather than a budget tier.
The economics shift matters for helloai's tracked set. Grok 4.3 was the cheapest frontier option at $1.25/$2.50 with 1M context — the cost floor that made it the Honest Daily Use leader. Grok 4.5 trades that for higher nominal rates and half the context window. Effective cost may still beat headline comparisons when output tokens dominate, because 4.5 burns far fewer tokens per resolved task. Teams routing on nominal leaderboard rates alone will mis-rank Grok against Gemini 3.1 Pro and Qwen3.7-Max until they profile output volume.
Availability is broad but not universal. Grok 4.5 is live in Grok Build, Cursor on all plans, and the xAI API console as of launch day. EU API access is delayed to mid-July per xAI's release notes. Grok 4.3 remains on the pricing table for workloads that need 1M context at the lower rate card. helloai tracks one xAI entry per admission rules — 4.5 wins because the provider docs steer new chat and code traffic there, not because 4.3 vanished.
The miss exposed a process gap, not a data gap. Memory listed rejected candidates like Grok 4.4; official catalogs moved faster. helloai now runs scripts/check_provider_catalog.py at the start of every weekly update — fetching each provider's catalog and news URLs and comparing parsed versions against models.json before judgment-heavy web search. The structural fix is boring and reliable: if docs.x.ai recommends Grok 4.5 and models.json still says 4.3, the pipeline stops until someone patches it.