Interestingly, Google no longer has a public frontier model. They have a very good flash model, but a very good flash model can't do frontier work without a good frontier orchestrator.
I am sure this will change soon, but Gemini 3.1 Pro is very clearly lagging at this point.
Positive users note Gemini 3.1 Pro's strong legal-task results while negative users criticize its slow reasoning, verbosity, missing releases, and Google's lagging frontier pace.
No Digg Deeper questions have been answered for this story yet.
A very good flash model may be all you need for mass serving consumers, but it doesn't do much for serious or agentic work.
Interestingly, Google no longer has a public frontier model. They have a very good flash model, but a very good flash model can't do frontier work without a good frontier orchestrator.
I am sure this will change soon, but Gemini 3.1 Pro is very clearly lagging at this point.
Its also a shame because Google has some of the most innovative AI apps, like NotebookLM, but they need smarter brains to power them (as well as the harnesses needed for those brains)
A very good flash model may be all you need for mass serving consumers, but it doesn't do much for serious or agentic work.
@emollick They said 3 weeks ago that Gemini 3.5 Pro was coming "in a month or a little bit more"
@emollick 3.5 flash is unnecessarily verbose and overexcited, not as expected
@emollick I'd almost call GLM 5.2 a frontier model
@emollick I am just confused why people expect Google to continue to enter this current rat race with OAI and anthropic. Wouldn’t it be smarter to just watch them fight from the sideline and not waste resources for now like they’ve done in the past? And deliver something big later
@deepanshusharmx Gemini models have always been weird.
The frontier orchestrator gap is real and the observation is correct. But domain-specific performance is a different story. Just published research today with a legal expert who co-wrote Slovenian legislation. We specifically tested Gemini 3.1 Pro on complex legal reasoning tasks. The results were striking enough that we called legal services the second biggest AI disruption currently underway, right behind coding. The "law is code" insight is literal, not a metaphor. Legal reasoning is variables, conditionals, dependencies, edge cases. Exactly what these models are optimized for. Flash for speed, Pro for reasoning depth on a specific domain. The absence of a general frontier model doesn't mean the model can't do frontier work in the right context. The question is whether Google understands how to position it or whether they're still chasing the general benchmark.
@emollick Gemini does great. None can solve disease. Humans will do that with AI. Its a team thing. Like James.
@emollick I'm not sure what is causing the delay, but judging from what they did with 3.5 Flash, it might be humongous scale of the new thing, so if they manage to complete the training/alignment it might be the new frontier.
@emollick Gemini 1.5 Pro's massive context window often compensates for reasoning gaps when used as a retrieval engine in complex agentic workflows.
I think gemini should focus on the doing more with less side rather then fighting in the frontier space. Make it the perfect digital assistant, research with google crawling mail writing sheets book keeping and more.
Not every model has to do everything but gemini has a strong lane
@emollick It’s a real bummer that Gemini 3 flash doesn’t seem like it will see a GA release either.
@emollick What is a flash model?
@emollick Bard vibes never left Googles AI
@emollick Gemini 3.5 Flash is closer to the frontier than Gemini 3.1 Pro
@emollick Gemini Flash is the intern without a senior
@emollick somebody on the gemini brand team is rewriting 'lagging' to 'iterating' for the monday keynote.
@emollick Wondering if this is why we're not seeing more healthcare-specific Gemini deployments yet. You need that frontier reasoning for medical complexity.
@emollick orchestrator routing bridges that gap anyway, chaining flash models with prompt caching usually handles vision tasks fine without the overhead