# Brood War Bench：Codex Astra、Claude Fable 与 Grok 4.6 等 Agent 的星际争霸对战评测 - 来源：Hacker News 热门（buzzing.cc 中文翻译） - 作者：benswerd - 发布时间：2026-09-20 06:55 - AIHOT 分数：55 - AIHOT 链接：https://aihot.news/items/cmu90fnbk03hxrojrb59kavfh - 原文链接：https://bw.swerdlow.dev/report ## AI 摘要 作者 bswerd 自建 Brood War Bench，让 Codex Astra、Codex 5.6、Claude Fable、Claude Opus 5 与 Grok 4.6 等配置在《星际争霸：虫群之心》中循环对战。 ## 正文 Key takeaways None of the models played beyond a beginner level. Codex Astra is the clear leader beating all other models consistently. Grok models are not smart enough to play Brood War yet. Older models tended to play the RTS as a turn-based game, leading them to get destroyed while they were thinking. Newer models sometimes fell into the same trap, which may explain why some lower-effort settings performed better, but overall were much more cognizant of the cost of thinking. Leaderboard RankSystemWinsLossesAPMCost / gameWin rate 🥇 Codex Astra / xhigh 18012.6$10.54100.0% 🥈 Codex Astra / medium 16217.2$15.1188.9% 🥉 Claude Fable 15312.6$12.2483.3% 4 Codex Astra / low 14425.7$21.0777.8% 5 Codex 5.6 Sol / medium 13510.1$5.1272.2% 6 Codex 5.6 Sol / low 12618.1$9.2366.7% 7 Claude Opus 5 12610.5$20.7866.7% 8 Codex 5.6 Sol / xhigh 1178.0$3.2361.1% 9 Codex 5.6 Luna / low 9923.8$0.4250.0% 10 Codex 5.6 Terra / xhigh 9915.8$2.1050.0% Brood War Bench started after I built a version of Brood War that you could only play through agents as an experiment to play with friends. I played it with a couple friends who did surprisingly well for people who have only played a couple Starcraft games in their lives. When I asked them why, they said they hadn't done much, they asked their agent to attack and it had built a small army and done the full attack for them. This lead me to wonder how far they can go on their own; this is my answer. What I observed 01 Codex found cheese before it found macro Codex's strongest recurring idea was disruption. In Protoss games it often sent a Probe across the map to attack workers or buildings. This worked shockingly well as the opposing agents often spent dozens of seconds thinking about what to do about a probe instead of doing anything else. The same systems were much weaker at sustained production. They delayed tech, trickled one or two basic units into defended bases, and threw workers into last stands. I also noticed Codex often created separate subagents to manage the economy, army production, and army control. They didn't communicate much with one another, so the army agent often sent each new unit straight into an attack, unaware of the larger army the other agents were planning to build. This is a common beginner mistake: sending units in one at a time instead of waiting for a critical mass and a planned attack timing. In games where I helped direct Codex, it was much better at planning those moments and getting its subagents to work together. The persistence was real. In G009, after losing its army and main base, Codex 5.6 Terra / medium lifted its last Command Center and moved it toward the opposite corner. It survived for another six minutes. 媒体内容 · 前往原文查看 Six Probes cross the map 媒体内容 · 前往原文查看 A Probe first, then Zealots in drips 媒体内容 · 前往原文查看 The last Command Center runs 02 Grok spent the game between actions Grok 4.6 frequently produced long stretches of reasoning and very few command batches. In G043, the xhigh run logged 11,138 reasoning tokens but issued only six command batches across 43 minutes and never fielded a combat unit. The actions it did take rarely developed into a working control loop. In G003, Grok / xhigh made three Marines and never reached the enemy base. In G002, Grok / medium made two Zealots and also never crossed the map. These looked less like bad strategies than failures to keep observing and acting. 媒体内容 · 前往原文查看 Forty-three minutes, no army 03 Fable earnestly tried to play the game I found myself rooting for Claude Fable in more than a few games. Fable usually tried to build an economy and climb the tech tree instead of stopping at the first unit available. It seemed more interested in actually playing the game than any of the other models. In G007 it reached a Lair, Spire, and Mutalisks and won. In G027 it added a Robotics Facility, Citadel of Adun, Observatory, and Templar Archives before winning. Ambition did not guarantee execution: in G036 Fable reached a Factory and Academy but Opus 5 overran it. 媒体内容 · 前往原文查看 Fable gets Mutalisks 媒体内容 · 前往原文查看 Fable keeps climbing 媒体内容 · 前往原文查看 The build does not become an army No agent here played beyond beginner level Even Astra and Fable were unable to build complex army's, defend simple attacks or play concrete strategies. A beginner playing photon rush would win every single one of these games. That said, watching the agents play made me more excited than I have been in a while. This benchmark is nowhere near exhausted. There is much more for the agents to learn, and much more for the benchmark to ask them to do. I look forward to watching them get there. How the games developed 媒体内容 · 前往原文查看 Technology investment Completed research + upgrade levels Codex Astra0.3Claude Fable0.2Grok 4.60 媒体内容 · 前往原文查看 Workers Completed workers alive Codex Astra14.7Claude Fable16.7Grok 4.67.4 媒体内容 · 前往原文查看 Army size Completed army and support units Codex Astra8.3Claude Fable7.6Grok 4.62 媒体内容 · 前往原文查看 Structures Completed buildings, including add-ons Codex Astra6.2Claude Fable7.4Grok 4.64.5 媒体内容 · 前往原文查看 Minerals in the bank Unspent minerals, not income Codex Astra240.6Claude Fable248.6Grok 4.6536.6 媒体内容 · 前往原文查看 Gas in the bank Unspent gas, not income Codex Astra151Claude Fable326.9Grok 4.6244.5 媒体内容 · 前往原文查看 Supply used Includes production in progress Codex Astra26.5Claude Fable26.6Grok 4.611.3 媒体内容 · 前往原文查看 When games ended Share of games ending per 5-minute window Time-series charts show means of recorded player-runs at each game time. Finished games drop out; missing samples are not filled. Models pool their effort settings. Units and buildings count only once completed; army excludes workers, Overlords, eggs, larvae, and ammunition. Win rate vs. cost Average cost per game, using the same prices as the leaderboard. Codex and Sonnet costs are token-based estimates. Codex Claude Grok How the benchmark ran We built a round-robin matrix of model and effort configurations and had every configuration play every other. The harness ran those matchups in parallel across Freestyle VMs, saving game-engine data and both agents' harness logs for each match. Head-to-head matrix Read across a row. W is a win, L is a loss, and T is a match that reached the benchmark time limit. Open the full 19 × 19 matrix W win L loss T time limit System12345678910111213141516171819 1Codex Astra / xhigh-WWWWWWWWWWWWWWWWWW 2Codex Astra / mediumL-WWWWWWWWWWLWWWWWW 3Codex Astra / lowLL-WWWWWWWWWLLWWWWW 4Codex 5.6 Sol / xhighLLL-LWWWLWWWLLWWWWW 5Codex 5.6 Sol / mediumLLLW-WWWLWWWLWWWWWW 6Codex 5.6 Sol / lowLLLLL-WWWWWWWWLWWWW 7Codex 5.6 Luna / xhighLLLLLL-WLWLWLLLWWWW 8Codex 5.6 Luna / mediumLLLLLLL-LLLLLWWWWWW 9Codex 5.6 Luna / lowLLLWWLWW-WLLLLLWWWW 10Codex 5.6 Terra / xhighLLLLLLLWL-WWLWWWWWW 11Codex 5.6 Terra / mediumLLLLLLWWWL-LLLWWWWW 12Codex 5.6 Terra / lowLLLLLLLWWLW-LLWWWWW 13Claude FableLWWWWLWWWWWW-LWWWWW 14Claude Opus 5LLWWLLWLWLWWW-WWWWW 15Claude SonnetLLLLLWWLWLLLLL-WWWW 16Claude HaikuLLLLLLLLLLLLLLL-LTT 17Grok 4.6 / xhighLLLLLLLLLLLLLLLW-WT 18Grok 4.6 / mediumLLLLLLLLLLLLLLLTL-W 19Grok 4.6 / lowLLLLLLLLLLLLLLLTTL- Play your own match Bring your agent and play Brood War with friends. Play Brood War