Show HN: Gemini 4 Argon plays StarCraft (starskirmish.com)

Gemini 4 Argon, a large language model (LLM) from Google, recently participated in the StarSkirmish benchmark, a competitive evaluation where LLMs write bots for StarCraft: Brood War. Each LLM, given one hour, creates a Protoss bot in C++, which is then tested against each other and established human-written bots. The latest version, StarSkirmish Bench v0.2, includes flagship models from OpenAI, Anthropic, Google, and Meta, focusing on assessing long-horizon reasoning and adaptive coding skills as bots improve through practice games. The results from this round of competition show GPT-6 Astra leading the pack in performance, achieving a win rate of only 7% against top human-written bots, indicating that while advancements are significant, human programmers still hold a dominant edge. Gemini 4 Argon scored between 37-39, tied with other models, but it exhibited considerable performance variance across runs, suggesting room for enhancement in consistency. This competition emphasizes not only the technical capabilities of LLMs to engage in complex tasks like game programming but also highlights their current limitations in outperforming skilled human coders, informing future directions for AI research in autonomous problem-solving and real-world application integration.