Grok 4.5 Closes In On Frontier

Grok 4.5 looks like a serious frontier contender, but the benchmark story is stronger for coding economics than universal capability.

• –Vendor-reported results place it near top models on software-engineering and terminal-agent evaluations
• –Independent measurements show a mixed picture, with Grok 4.5 competitive but trailing the best scores on several coding benchmarks
• –Its $2 per million input-token and $6 per million output-token pricing makes long-running agent workflows unusually affordable
• –Joint training with Cursor and real developer-agent interaction data may explain its strength on practical coding tasks
• –Developers should test it against their own repositories, since benchmark harnesses and model-serving conditions can materially change results