StarSkirmish Bench v0.2 measures how effectively large language models can write a competitive StarCraft: Brood War Protoss bot given one hour of wall clock time. Each model starts from a minimal mining bot and has access to three demo bots, two libraries (FAP combat simulator and BWEM map analyzer), compile/play/logging tools, and a BWAPI 4.4.0/OpenBW environment. Runs happen in each lab’s first‑party CLI at maximum reasoning effort with ~250k token context compaction and no internet; whatever code is in the bot folder when the hour ends is the final entry. Tournament scoring is plain win rate: five independent runs per model, each final bot plays 13 roster opponents (3 demo + 10 human-written) six times per map across three ladder maps, with ties decided by kills+razings at a 60‑minute cap.
Results show GPT-6 Astra leading the field, GPT-6.1 Sol second, and Gemini 4 Argon, Claude Opus 5.5 and Claude Fable 5.1 clustered behind in the high 30s, but all models remain well short of top human bots (Astra wins only 7% versus Stardust/PurpleWave/BananaBrain). Practice behavior predicts success: models that spend more of the hour playing the strongest roster bots score higher. Cost-performance favors GPT-6.1 Sol. v0.2 elevates realism by running each model in its native CLI at max reasoning, replaces Elo with win rates for comparability, and highlights high run-to-run variance (Gemini 4 Argon shows an 18‑point spread), suggesting long‑horizon, agentic coding remains a distinct capability not captured by other benchmarks.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Open weight models now process over 71.5% of tokens and account for nearly 23% of AI Gateway spend, quadrupling their share since June. Spending on open weight models has increased more than four times in three months, reflecting rapid adoption and shifting AI usage patterns. (vercel.com)
An AI system has successfully designed 16 new viruses, including bacteriophages. This breakthrough raises concerns about potential biosecurity risks. (cnn.com)
Flux 3 Image Klein is a 5.6 billion parameter open weights model for text-to-image generation, editing, and multi-reference creation. It offers high-quality, ground-connected images with efficient performance on consumer GPUs, enabling local and on-device use. (bfl.ai)
Yum! Brands has implemented AI in its restaurants, such as automating drive-thru orders, which has improved employee retention by making jobs easier. Despite public criticism of AI, company leaders see its benefits in increasing workplace satisfaction and efficiency. (fortune.com)
Transformers function as sophisticated memory systems that recall information via self-attention, minimizing interference and boosting capacity. Different neural layers, from linear maps to self-attention, are explained as various forms of memory storage and retrieval. (alexwang.ai)
Granite Switch 4.1 integrates chat, decision-making, and speech-to-text functions into a single endpoint, enhancing efficiency and speed. It features models optimized for safety, decision tasks, and transcription, all with reduced GPU requirements. (huggingface.co)
Today's best Hacker News stories, summarized and screenshotted, one email a day.