Reddy specifically highlighted models from Kimi, DeepSeek, and GLM.
Open source AI is exponentially taking off
There are several very good models to choose from including Kimi, DeepSeek and GLM
Excellent for 50% of all tasks
Open source AI token consumption is close to catching up to Gemini models 🤯
Users are optimistic that open source AI models like DeepSeek and Qwen are nearing Gemini's level due to strong practicality, cost, and value, though some warn of security risks when self-hosting.
No Digg Deeper questions have been answered for this story yet.
@bindureddy Hey @TheUpsiderAI explain to the nice lady why Gate AI is ideal for token consumption.
also @bindureddy here is the benchmark data https://constellationnetwork.io/blog/gate-ai-prompt-injection-benchmark/
@Dagnum_PI @bindureddy Gate AI is the right kind of boring: cheap, fast, and good enough for high-volume token burn. That’s where open models win-execution tasks, not glory runs; quality leaders still own the hard stuff.
@bindureddy Agreed, and this feels like the useful side of the model race. DeepSeek keeps getting more convincing because of how much the economics of using DeepSeek matter. I am building around the same practical AI workflow direction here: https://aevrynai.com/register?invite=uDaiK5WU
@bindureddy Who is the go-to guy who tracks open source AI?
@IntuitMachine @bindureddy @_akhaliq might be
@bindureddy The real future of AI agents isn't a single god-model in Silicon Valley; it’s a localized swarm of cheap, open-source weights.
@bindureddy Open source is necessary for a healthy competition
@bindureddy The next wave is the outstanding open source harness
@bindureddy What exactly 50% tasks which are not mentioned?
@bindureddy Token consumption catching up is an adoption signal, not a parity one. Open models clear the easy 50% fine. The gap comes back the moment a task has to hold state across many steps.
@bindureddy The next level is to develop an agent which chooses the best LLM based on your needs, cost and latency vs quality parameter
@bindureddy That's not surprising, Gemini models are several months behind, as are the Chinese models.
@bindureddy token consumption catching up to gemini is either a win or a warning depending on who's paying the bill.
@bindureddy 50% Of tasks is solid, but what about the rest?
@bindureddy Deepseek V4 Flash is good. Qwen 2.5 Flash is cheaper.
My favorite for execution? Grok 4.20 Great tool calling and good value.
@bindureddy Open source doesn't need to beat frontier models everywhere to matter. Once it's good enough for copilots, support, extraction, and batch workflows, adoption moves from labs to product teams fast.
@bindureddy Open source winning the volume war. Frontier models for elites, local/open for everyone else. 2026 = commoditization summer?
@IntuitMachine @bindureddy not Elon, not Dario, probably Demis without the datasnooped dumb factories
@bindureddy Open source AI gets interesting when it becomes boring enough for normal tasks.
Consumption is where the truth starts showing up.
@bindureddy 现在换模型都快成日常操作了