Please don't discontinue Gemini 2.5 Flash (discuss.ai.google.dev)

Article summary

The author is requesting that Google not discontinue the Gemini 2.5 Flash model, as it performs well for their specific workflows and is more cost-effective than the newer Gemini 3 Flash model. The author's benchmarks show that Gemini 3 Flash does not perform as well as Gemini 2.5 Flash, even with tweaks to prompting. The author believes that many users rely on Gemini 2.5 Flash and would be negatively impacted by its discontinuation. The model's low latency and performance make it suitable for voice agents and other applications.

Main themes

• Model discontinuation
• Performance comparison
• Cost-effectiveness
• Workflow disruption
• AI model upgrades
• Cloud dependency

What commenters say

• The discontinuation of Gemini 2.5 Flash would be a significant loss for users who rely on its low latency and performance for specific applications.
• Newer models like Gemini 3 Flash are not necessarily better than older models like Gemini 2.5 Flash, and may even perform worse in certain tasks.
• The constant upgrading of AI models can be problematic for developers who build their workflows around specific models, requiring them to rebuild and retest their systems frequently.
• Relying on proprietary models can be risky, as they can be discontinued or changed without notice, and using open-source models can provide more control and stability.
• The pricing of newer models is a significant concern, with some users feeling that the cost increase is not justified by the performance improvements.
• Some commenters argue that developers should be prepared to adapt to changing models and APIs, while others believe that this can be a significant burden and a barrier to adoption.
• The use of cloud-based models can be convenient, but it also means that users are at the mercy of the provider and may not have control over the models they rely on.
• There are alternative models available, such as open-source models, that can provide similar or better performance than proprietary models like Gemini 2.5 Flash.