今天新增gemini-4IMAGE
LeanLeanBench：Gemini 4 Argon 输出 token 消耗偏高
Opus 5.5、Sol 6.1 与 Gemini 4 Argon 同处帕累托前沿，但 Opus 读取更省，Argon 需更多输出 token 才达到同分。
Kári Rögnvaldsson
@karirogg8 小时前收录原推发布于 2026.10.28
Kári Rögnvaldsson @karirogg原推发布于 2026.10.28
Gemini 4 Argon now on LeanLeanBench! Opus 5.5, Sol 6.1, and Gemini 4 Argon now all lie pretty much on the Pareto frontier, but it turns out that Opus is much more read-efficient than the other models and Gemini Argon uses a lot more output tokens to reach the same score.
x.com/karirogg/status/2115137780863131661采集于 2026.10.28 10:03