Grok 4.5 Goes Public Tomorrow: xAI Undercuts Opus at a Quarter of the Output Price
TL;DR
xAI opens Grok 4.5 to the public on July 9 at $2/$6 per million tokens, a quarter of Opus pricing, but skips the EU and independent benchmarks.
Elon Musk said on July 8 on X that SpaceXAI will open Grok 4.5 to the public on July 9, pitching it as "an Opus-class model, but faster, more token-efficient and lower cost." It is the model's first release outside the SpaceX and Tesla private beta that started in late June.
The base is a 1.5-trillion-parameter V9 foundation model, blended with training data pulled from real Cursor IDE coding sessions. It is the first public output of xAI's acquisition of Anysphere, Cursor's parent, which Musk described as putting "the model lab, the IDE interaction data and the compute under one roof."
Pricing is aggressive. Grok 4.5 costs $2 per million input tokens and $6 per million output tokens, with three configurable reasoning-effort tiers (low, medium, high; default high). Anthropic's Claude Opus 4.8 currently lists at $5 input and $25 output. Grok 4.5's output price is 24% of Opus. Launch channels include Grok Build, the SpaceXAI console and Cursor across all subscription tiers.
The EU is not in the launch list. Musk gave no timeline, saying only that it will ship "after the compliance review."
The release lands the same day as OpenAI's GPT-5.6 Sol public launch and Anthropic's Fable 5 limited-access rollout, three frontier drops in a single window. Musk went after Opus by name in the post and left the GPT camp alone: "Beta customer feedback has been very positive, early evals are at or above Opus."
No independent benchmark numbers are out. Analysis site explainx.ai wrote that "Opus-class" is a marketing claim until community evals appear. TechTimes reported on June 29, during the SpaceX and Tesla private beta, that xAI had not disclosed MMLU or SWE-bench scores.
Musk is betting Grok 4.5 on coding while pricing it four times below the model he keeps naming. To prove a cheap Opus exists, xAI first has to let developers run the benchmarks.
via Axios / Bloomingbit / ExplainX / TechTimes
The base is a 1.5-trillion-parameter V9 foundation model, blended with training data pulled from real Cursor IDE coding sessions. It is the first public output of xAI's acquisition of Anysphere, Cursor's parent, which Musk described as putting "the model lab, the IDE interaction data and the compute under one roof."
Pricing is aggressive. Grok 4.5 costs $2 per million input tokens and $6 per million output tokens, with three configurable reasoning-effort tiers (low, medium, high; default high). Anthropic's Claude Opus 4.8 currently lists at $5 input and $25 output. Grok 4.5's output price is 24% of Opus. Launch channels include Grok Build, the SpaceXAI console and Cursor across all subscription tiers.
The EU is not in the launch list. Musk gave no timeline, saying only that it will ship "after the compliance review."
The release lands the same day as OpenAI's GPT-5.6 Sol public launch and Anthropic's Fable 5 limited-access rollout, three frontier drops in a single window. Musk went after Opus by name in the post and left the GPT camp alone: "Beta customer feedback has been very positive, early evals are at or above Opus."
No independent benchmark numbers are out. Analysis site explainx.ai wrote that "Opus-class" is a marketing claim until community evals appear. TechTimes reported on June 29, during the SpaceX and Tesla private beta, that xAI had not disclosed MMLU or SWE-bench scores.
Musk is betting Grok 4.5 on coding while pricing it four times below the model he keeps naming. To prove a cheap Opus exists, xAI first has to let developers run the benchmarks.
via Axios / Bloomingbit / ExplainX / TechTimes
