xAI Launches Grok 4.5 to Undercut Anthropic and OpenAI
2 min readxAI publicly released Grok 4.5 on July 8, and the headline is not a benchmark record. It is the price tag. Elon Musk’s AI lab is selling its newest coding and agentic model at $2 per million input tokens and $6 per million output tokens, undercutting Anthropic’s flagship Opus 4.8 while claiming to sit in the same performance class.
The Frontier Race Gets a Price War
For most of the past year, the competition between frontier AI labs has been fought on capability. Anthropic, OpenAI, and Google have traded the top spots on coding and reasoning benchmarks, and pricing largely followed performance. Grok 4.5 flips that script by leading with economics.
The model ships with a 500,000-token context window, which is actually smaller than the million-token window listed for the earlier Grok 4.3. xAI is signaling that its priority this release is cost-efficient real work rather than maximal context.
What xAI Actually Shipped
On xAI’s own benchmarks, Grok 4.5 beats Opus 4.8 on the provider-run DeepSWE 1.0 score and on Terminal Bench 2.1, but trails Opus 4.8 on the neutral DeepSWE 1.1 run and on SWE Bench Pro. Independent evaluator Artificial Analysis places Grok 4.5 at 54 on its Intelligence Index, ranking it fourth overall rather than first.
The more interesting claim is efficiency. xAI says Grok 4.5 uses roughly 4.2 times fewer output tokens than Opus 4.8 to finish comparable SWE Bench Pro tasks. If that holds up, a job that runs cheaply on Grok could cost several times more on a rival even at similar per-token prices. The model was trained alongside Cursor and ships in Grok Build, Cursor, the xAI API, and Microsoft Office add-ins.
Why It Matters
Grok 4.5 reframes the model race around dollars per finished task rather than raw benchmark peaks. For developers and companies running agents at scale, cost per completed job often matters more than a single leaderboard number. The catch is that xAI’s strongest numbers come from its own harness, so independent verification will decide whether the Opus-class claim is marketing or reality.
Watch for neutral third-party benchmarks in the coming weeks, and for how Anthropic and OpenAI respond on price.