SpaceXAI releases Grok 4.5, an 'Opus-class' model that undercuts rivals on price

Grok 4.5 shipped inside the same 72-hour window as GPT-5.6 and Meta's Muse Spark — the first time multiple frontier models landed essentially simultaneously — and its pitch is aggressive pricing rather than a raw capability crown. Musk called it an 'Opus-class model' that dramatically undercuts Anthropic and OpenAI, and told Tesla staff to switch to Grok internally, with engineers at Tesla and SpaceX reportedly finding it highly useful.

The most credible external validation came from Perplexity CEO Aravind Srinivas, who said Grok 4.5 scored highest on Perplexity's internal WANDR agentic-research benchmark 'inside the Computer harness,' at half the price of Claude Opus 4.8 (high) and even ahead of Perplexity's GLM 5.2 post-train (1,544 likes). The model was developed in collaboration with Cursor and touts stronger real-world engineering performance and faster token throughput.

Musk himself tempered expectations on coding supremacy, posting that 'Grok places second after Fable on real-world software engineering' (8,370 likes, 1,431 retweets) — a candid admission that Claude Fable 5 still leads on SWE work. That fits the week's pattern: the differentiator is dollars-per-task, not benchmark bragging rights. The launch also lands days after xAI's corporate rebrand to SpaceXAI, folding all AI products under the SpaceX umbrella. Watch whether the price advantage pulls agentic workloads away from Anthropic and OpenAI, or whether Fable's edge on production SWE-Bench Pro keeps serious coding teams on Claude.