Gemini 3.8 Live and 3.8 Live Extended Thinking

Post by asdfman123 » Wed, Sep 16, 2026, 7:55 PM UTC

Gareth321 wrote: post #246

rdtsc wrote: post #199

Earlier quoted context omitted.

I think it's good at basic coding and is cheaper and faster. Trying it vs Fable 5.1 I don't see it being even close capability-wise. What I mean by basic coding is "write a script to parse this data" and answering some questions based on the code. It does that well and answers fast. However, when it comes to planning and thinking through a more complicated design problem it's just not there yet.

It's not even close. Some benchmarks show 3.8 Flash as being close to Astra (DeepSWE v1.1), but they are mostly "bench-maxed". Newer benchmarks like Terminal-Bench 4.0 show Astra at 57.9% and 3.8 Flash at 19.1%. For anyone who has tried to code with Gemini models, there is no contest here. Astra wins handily on every other benchmark, too. Research, science, 3D, visual, Humanity's Last Exam, etc.

Post by MoreThanMe » Wed, Sep 16, 2026, 9:03 PM UTC

thisgoodlife wrote: post #110

phenomen wrote: post #54

Earlier quoted context omitted.

It's also the only model that generates accurate translation and localization. No other frontier model comes close. Although Gemini's coding capabilities are subpar, its natural language processing is top-tier.

I’m curious how you guys keep track of each model’s coding capabilities. The landscape keeps changing. I don’t suppose you benchmark all frontier models every other month, right?