Gemini 4 Carbon 在 Bug Bunt Bench 得分 57.3，成本仅 Fable 5.2 的四分之一
有人拿 105 个真实 bug 实测了几家旗舰模型，Gemini 4 Carbon 只差 0.7 分追平 Fable 5.2，花费却只有 $31.32 对 $139.65。做代码修复选型的人应该看看这份带价格的榜单，比官方宣传单靠谱。
独立测试者在 Bug Bunt Bench 上用 2 个仓库、105 个重新植入的真实 bug 测评了多款旗舰模型，Gemini 4 Carbon (high) 拿到 57.3 分，仅以 0.7 分之差落后于榜首的 Fable 5.2 (max)。成本差距更明显：Fable 5.2 花费 $139.65，Gemini 4 Carbon 只花 $31.32，约为前者的四分之一。
对比其余选手：Sonnet 5.5 (max) 得 51.3 分但花费 $154.03，GPT-6 Astra (max) 得 45 分，GPT-6.1 Sol (max) 得 44.3 分（仅 $5.88），上一代 Gemini 4 Argon (high) 得 40.7 分。Carbon 相比 Argon 提升 +16.6 分，与 Google 官方基准宣称的数字一致。
整个测试通过 Antigravity CLI 执行，属于真实代码库上的修复任务而非合成评测。对开发者而言，这份测评把“找 bug 并修复”的能力和 API 花费放在一起比较，是选型时少见的性价比参考。
Insane jump from Argon to Carbon on Bug Bunt Bench using Antigravity CLI. Crazy how fast we are moving and how much more cost efficient the Gemini models are for the capability! Paweł Huryn @PawelHuryn What a day! I tested Gemini 4 Carbon on real work. 2 repos, 105 replanted bugs. Find and fix what you can. It scores just behind Fable 5.2. The results: - Fable 5.2 (max): 58 for $139.65 - Gemini 4 Carbon (high): 57.3 for $31.32 - Sonnet 5.5 (max): 51.3 for $154.03 - GPT-6 Astra (max): 45 for $33.03 - GPT-6.1 Sol (max): 44.3 for $5.88 - Gemini 4 Argon (high): 40.7 for $7.16 Google's own benchmarks looked super strong, but nobody could access the model. Turns out they didn't lie: +16.6 points over Argon, at less than a quarter of Fable 5.2's cost. Nobody should be laughing at Google anymore. (the next effort level drops soon in the thread 🧵) 🔗 View Quoted Tweet📊 9 ⚡
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison and AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Gemini 4, Bug Bunt Bench, Fable 5.2, Antigravity CLI.
Source attribution: curated from Varun Mohan (@_mohansolo). Cross-source corroboration: 11 independent sources have covered this development (including berryxia, IT之家, 宝玉), confirming industry-wide signal strength.
