@SpaceXAI announced Grok 4.5 โ purpose-built for coding, knowledge work and agentic tasks. Trained on NVIDIA GB300 NVL72.
In early benchmarks, Grok 4.5 scores 83.3% on Terminal-Bench 2.1, 78.0% on SWE-Bench Multilingual, 62.0% (high) on DeepSWE 1.0, and 64.7% (high) on SWE-Bench Pro. It shows strong performance across both terminal and software engineering tasks.
This release positions Grok 4.5 as a highly capable model for real developer workflows, combining solid reasoning with practical coding abilities.
Announcing Grok 4.5, our first model trained specifically for coding and agents. It was trained with Cursor and offers frontier intelligence at leading speeds and cost efficiency.
https://t.co/i8HpU7w64k
@blknoiz06 Markets tend to move before everyone feels comfortable buying. If everyone is waiting for the same window, that window rarely stays open for long
@galshirart I don't think AI is replacing designers - its replacing the parts of design that were limited by execution speed. Taste, direction, and knowing what actually matters are still the hard problems
@AlexFinn I think the biggest shift isn't even the models themselves. It's that AI is becoming something you work with all day instead of something you occasionally open to ask a question.