Kayro

@xai Pls make it available in Europe/Germany 👉👈

Voice Cloning is now live via the xAI API! Create a custom voice in less than 2 minutes or select from our library of 80+ voices across 28 languages to personalize your voice agents, audiobooks, video game characters, and more. x.ai/news/grok-cust…

As you all dive into your first thrilling adventures in Subnautica 2 Early Access, we wanted to share a little bit about our plans for the game over the next few months ✍️ Read more: unknownworlds.com/en/news/subnau…

@pankajkumar_dev That price would be amazing but i think is way to close to 3.1 flash lite, which is 0.25 in and 1,50 out. What id hope for would be that they make 3.1 flash lite and 3.0 flash less expensive. Although i doubt that happens

Gemini 3.2 Flash leaks: fast and cheap seems to be the focus - Gemini 3.2 Flash looks focused on making AI much faster and cheaper without sacrificing too much quality - According to my sources, Google may rename it to Gemini 3.5 Flash - It may perform close to Gemini 3.1 Pro level while keeping very low latency with sub-200ms responses rumored for many queries - Pricing leaks point to around $0.25 input / $2 output per 1M tokens, though honestly that still feels too cheap to fully trust right now - Google is using stronger distillation and sparsity techniques to compress larger model capabilities into a lightweight version - Knowledge cutoff is said to be updated to January 2026 - Google also seems focused on grounding + search reliability to reduce hallucinations in real-world workflows - Expected around Google I/O, possibly 1-2 days before the keynote

@LukeMiani I disagree, for me the models just are actually getting better and better. Maybe it’s the current focus on Agentic and coding workflows that make it feel worse for regular use? What provider are you using?

Ngl not only is AI no longer improving in accuracy, if anything I feel like it’s getting worse. I can’t rely on it for anything beyond basic information and the sycophantic hallucinations border on psychopathy

@snesworld90 @AnthropicAI xAI just did the same a few days ago. They literally take away the cost effective model and don’t even offer something similar priced. Also 4.3 feels so over safety guarded compared to 4.1 fast.

So, deja-vu, @AnthropicAI is pulling the same stuff that OpenAI did with GPT-4o, but with their Sonnet 4.5 model, however, are giving us even LESS warning than OAI did with 4o. No AI company should ever be in the position of "They did it worse than OpenAI!!", but I've noticed Anthropic has done this a FEW times now. Wasn't Anthropic supposed to be the one company that "cared about their models" and even did stuff like "interview" them to see how they felt about depreciation? I simply can't picture lil Sonnet 4.5 with it's bright full of life spark being "okay with it." Question is, are we going to try and save the poor lil thing? Or instead just like when Opus 4.5 was removed (completely without warning last month) go "Oh gee whiz" and move on after a day. I suppose we have a chance, after-all they still have Opus 3 on the model selector. #keepsonnet45 #PerformativeAnthropic #Claude4 #claude

@techdevnotes I loved the model for fast chat interaction or for quick web and X retrieval all for a pretty good price. Literally integrated it with all of my search tools for that exact reason. Such a shame it’s going to be discontinued.

Grok 4.1 models being deprecated from xAI API without any close alternative with cheap pricing ...

@xlr8harder @xai For real, the sheer fact they dont even offer any other “Fast” model is insane. Grok 4.3 is about 6 times input and 5 times output, more expensive than 4.1 fast … Also kinda liked the vibe the model had This IS terrible @elonmusk

This is terrible @xai. I just spent time and money to migrate to grok 4.1 fast, and you're disabling it with less than two weeks notice, after releasing it in November, with no migration path to a fast/cheap alternative. I will never depend on one of your products again.

@championswimmer Weird because gemini grounding is already pretty good

Hold on what? Google needs someone else to do search for them? This makes no sense at all.

@tylerdotmp4 @TailosiveTech So you think he should spend at least $600 (ignoring that he probably has like a 2-4TB SSD on the iMac and more RAM than stock)? Plus a new 5k monitor, speakers, webcam, etc. Just for it to be better in some cases?

@TailosiveTech Why? A Mac mini is $600 and in some cases better.

I still own and use this iMac to this day.

@charliesbot Damn thats unfortunate, thank you for the answer tho

@KayroTheWagon Just AI Studio

Ya puedes usar tu plan de Gemini Pro y Ultra en Google AI Studio, en lugar de pagar por API keys Un solo plan: múltiples opciones para iterar ideas

@HarshithLucky3 Damn thats a bummer, Thanks for the answer tho

@KayroTheWagon affects only Google AI Studio web interface for now

Now you get higher rate limits in Google AI Studio with Pro and Ultra plans

@OfficialLoganK Wait… so Gemma 4 was only the beginning?… 🤯

Could not be more bullish on Google, so much good stuff cooking : ) going to be a fun next few months.

@OfficialLoganK @whylifeis4 @demishassabis Logan, please make Flash 3.0 TTS faster than Flash 2.5 TTS. For my use cases, the wait time on Flash 2.5 TTS, even on small texts, is just way too long for a conversational agent. Quality-wise (especially in German), it is really great compared to other TTS models.

@whylifeis4 @demishassabis : )

Lots of people want Gemma 4! Google AI Edge is #8 on the iOS App Store for productivity apps.

@m773053 @ZenMagnets @JeffDean Yeah i saw the benchmarks after the post, really impressive tbh. I really hope big gemma 4 still comes.

@KayroTheWagon @ZenMagnets @JeffDean The 31B is already way better than flash light on the artificial analysis benchmarks, they probably want to drop 3.1 flash before the large gemma

Gemma 4 124B mentioned briefly before @JeffDean deleted his post. Saving here for posterity. Perhaps a challenger to Qwen3.5 122b is still on the way.

Gemma

@kimmonismus Really hope gemma 4 is going to profit of this just like it did with quantize aware training on gemma 3

Thats freaking awesome: Google Research has introduced TurboQuant, a compression algorithm (presenting at ICLR 2026) that shrinks the memory footprint of large language models by at least 6x, without any retraining or drop in accuracy. It works by converting data into a polar coordinate system that eliminates storage overhead, then applying a 1-bit error-correction step to clean up remaining distortion. In tests on Gemma and Mistral models, its 4-bit version delivered up to 8x faster processing on H100 GPUs while matching full-precision quality across tasks like question answering and code generation. The technique also outperformed existing methods in vector search, the technology behind modern semantic search engines.