Gemini 4 Neon 成为 Models Engineering Index 的新第一。
Gemini 4 Neon is the new #1 on the Models Engineering Index

原始链接: https://git-aarvin.github.io/mei/

该文本比较了19个AI模型的指数分数、提供商以及相对成本/令牌用量。Neon排名最高(69.5),其次是Fable 5.2、Opus 5.5和6.2 Astra;Luna排名最低(34.8)。它还显示了相对于基线的成本和令牌倍数,并比较了选定模型在包括CursorBench、FrontierCode、Code Migration、IOI、ProgramBench、Terminal-Bench和Vibe Code等编码/智能体基准测试中的表现。基准分数因模型和任务而差异很大。数据归因于Cognition、Cursor、Laude Institute和Vals AI,来源在GitHub。

Gemini 4 Neon 在 Models Engineering Index 排名第一。评论者质疑信任单一指数,建议独立指数可能较少受刷榜影响,并批评 Google 模型在编程方面表现不如 Claude/OpenAI,且 Token 使用量高、幻觉率高。
相关文章

原文

.

30405060704 Neon69.5Fable 5.265.8Opus 5.563.16.2 Astra60.54 Hydrogen57.2Sonnet 5.557.26.2 Sol56.9Wonder Full49.3K3.1 Pro49.04-Max49.06.1 Terra47.2Wonder46.4V4.1 Flash44.1Grok 4.742.7Haiku 5.542.5GLM 5.342.0Hy437.3V2.6-Pro35.46 Luna34.8 8×4×2×1×0.5×0.25×0.125×0.0625×6 Luna0.065×V4.1 Flash0.13×V2.6-Pro0.13×Hy40.33×Haiku 5.50.4×6.2 Sol0.4×6.1 Terra0.83×Wonder Full1.1×K3.1 Pro1.2×4 Hydrogen1.2×Wonder1.3×6.2 Astra1.4×4 Neon1.9×GLM 5.32.2×Sonnet 5.52.5×Opus 5.53.5×Grok 4.73.9×Fable 5.26.5×4-Max7.3× 2×0.5×6.2 Astra0.29×6.2 Sol0.35×Wonder Full0.49×6 Luna0.62×Wonder0.63×K3.1 Pro0.74×Grok 4.70.9×V2.6-Pro0.93×4 Neon1×4 Hydrogen1.1×Hy41.1×4-Max1.2×V4.1 Flash1.4×GLM 5.31.5×Haiku 5.51.6×Opus 5.51.7×6.1 Terra1.7×Fable 5.22.1×Sonnet 5.52.5×

Performance versus .

30405060708×4×2×1×0.5×0.25×0.125×0.0625×4 NeonFable 5.2Opus 5.56.2 Astra4 HydrogenSonnet 5.56.2 SolWonder FullK3.1 Pro4-Max6.1 TerraWonderV4.1 FlashGrok 4.7Haiku 5.5GLM 5.3Hy4V2.6-Pro6 Luna 30405060702×0.5×4 NeonFable 5.2Opus 5.56.2 Astra4 HydrogenSonnet 5.56.2 SolWonder FullK3.1 Pro4-Max6.1 TerraWonderV4.1 FlashGrok 4.7Haiku 5.5GLM 5.3Hy4V2.6-Pro6 Luna
Index bar
14 Neon Google69.51.9×1×
2Fable 5.2 Anthropic65.86.5×2.1×
3Opus 5.5 Anthropic63.13.5×1.7×
46.2 Astra OpenAI60.51.4×0.29×
54 Hydrogen Google57.21.2×1.1×
6Sonnet 5.5 Anthropic57.22.5×2.5×
76.2 Sol OpenAI56.90.4×0.35×
8Wonder Full Meta49.31.1×0.49×
9K3.1 Pro Moonshot49.01.2×0.74×
104-Max Alibaba49.07.3×1.2×
116.1 Terra OpenAI47.20.83×1.7×
12Wonder Meta46.41.3×0.63×
13V4.1 Flash DeepSeek44.10.13×1.4×
14Grok 4.7 xAI42.73.9×0.9×
15Haiku 5.5 Anthropic42.50.4×1.6×
16GLM 5.3 Z.ai42.02.2×1.5×
17Hy4 Tencent37.30.33×1.1×
18V2.6-Pro Xiaomi35.40.13×0.93×
196 Luna OpenAI34.80.065×0.62×

Comparing and

255075100CursorBenchFrontierCodeCode MigrationIOIProgramBenchTerminal-BenchVibe Code 77.872.879.663.377.374.280.871.579.9
联系我们 contact @ memedata.com