Bindu Reddy is an AI commentator who posts on X about model capabilities, pricing and where the industry is heading. Recent posts focus on Google's newly announced Carbon model, including his LiveBench expectations, pricing speculation and a rumored OpenAI model called Bel, plus his purchase of Google shares. He also covers Argon's weak instruction-following, Chinese models distilling Astra, Anthropic's reliance on API revenue, and AI moving from coding to physics.
AI is hitting a human wall which will cause the bubble to burst 💥 Most humans and companies that they work in don’t view AI favorably They actively restrict AI usage in their organizations and at best view it a $20/month employee perk They are definitely not…
🚨 AI trends in the coming months - GPU demand will go up - token prices will do down - programming jobs will begin to disappear - some Neo clouds will go bankrupt - inevitable consolidation in the market place - Chinese models will have double digit market s…
🚨 Train Your Own LLM Using A Text Prompt - start with a text prompt - customize based on your use-case - run LLMs locally Try it on Abacus AI SuperComputer
Age of Abundance - How work becomes optional - AGI for all - mass market robotics, white-collar automation - super intelligence solves for near-infinite energy - gambling and investing will be prevalent - entertainment will be huge - taxes should be hig…
Woot! Now that Google is launching Carbon OpenAI will also launch Bel, their big internal model This is great news! AGI needs to be set free 🚀
X is extremely biased towards coding models - Argon excels at data analysis - Astra is very good at browser use Both models are vastly underhyped because of a few vocal developer and influencer accounts And everyone just echoes each other for engagement :(
Just bought a chunk of Google shares They will do insanely well in the AI era given that they have compute and frontier models! Plus Carbon just got announced Buy GOOG! 💥🚀
Jon Stewart is normally on point but it’s sad to see him be so utterly clueless about AI The AI revolution is not a house of cards or bunch of circular deals, It’s completely transformed white collar work, media and software Work will be optional and humans w…
It’s unbelievable how many famous people are using AI to tweet Suddenly, X seems so much more smarter, insightful and full of witty rebuttals LOL! It’s not the famous humans with followers, it’s their AI agents speaking in their tone 😅
The Chinese models have moved on to copying Astra! They are now actively distilling from Astra because Anthropic blocked them Over time all the models will smell like Astra and the harnesses will optimize for OpenAI Plus OpenAI gets 6-7% distillation revenue!
Gemini Carbon will be on LiveBench later today! Super excited to see where it lands.... I don't think it will be #1, but it will be competitive
Love it! Google just announced Carbon It's not available yet, but it will almost definitely be comparable to Fable and Astra It could cause a mini-earthquake if it's priced the same as Argon and is 5x cheaper
80% of Anthropic's revenue comes from API consumption pricing This is their soft spot. DeepSeek costs 20x less than Opus and works for 70% of the tasks already Companies are massively overspending if they use expensive tokens for the majority of simple tasks
LiveBench finally reflects real-world performance - Fable 5.1 and Astra are on top - Opus 5.5, not far behind, excels at design - Argon beats Terra, good at data analysis - Google beats open-source for the first time - DeepSeek is the top open-source model
AI has pretty much solved coding The next big frontier is physics. The math problems that AI is cracking right now is a stepping stone to physics A seminal physics breakthrough can lead to solving for near free infinite energy! Of course, this would automatic…
Argon is bad for coding because it's not very steerable It's bad at instruction following, which annoys devs, and hallucinates at times Both are simple bug fixes
We will have at least a dozen AI models that can do all mundane white collar work in six months Yes that includes coding, shopping and writing docs Models are already becoming a commodity and this trend will accelerate dramatically
Astra 6.1 drops as soon as next week At $30 per 1M output tokens, it will beat Opus 5.5 Astra has way better vibes than Opus, and humans love chatting with it
🚨 Announcing Smaug Humanizer - Humanize your AI-generated text in your tone - open-weights dropped on HF - plug-and-play to humanize your text - sound more human and less robotic - built on Gemma Destroy AI-slop and bring creativity back to the internet
LOL! GPUs are long lasting than models GPUs can last up to 6 years while model weights become legacy in 6 months Today, it’s Fable and Astra, tomorrow it will be Bel and Carbon at a fraction of the price
🚨 GPT 6.1 Terra Is The Cheapest Frontier Model Per Task - cheaper than other frontier models - comparable to Argon and Sonnet in total score - very good at browser use A really good fit for personal agents and as a chatbot
GPT 6.1 Terra Just Launched With Mind-Blowing Pricing $1 per 1M input $5 per 1M output Terra tends to be a Sonnet-class model and is now HALF the price of Sonnet. 🚀
Anthropic is releasing their financials in a few minutes All the AI bubble talk will vanish after seeing their revenue pop TBH the more interesting number will be their cost of training. I expect that loss to be substantial
PREDICTION - Google overtakes OpenAI and potentially Anthropic by year end Swapping all our Sol and Sonnet workloads with Argon. It’s just better Their next release Carbon will beat Fable and Astra. It almost seems inevitable 😲
US open source is insanely behind Chinese models All the research labs that got funded billions to deliver AI models have fallen short Apparently the good news is that they can resell the GPUs they bought for research on spot markets! Yikes 😳
Deepseek Flash is very hard to beat on easy tasks - ship updates every month - 10x cheaper than sonnet - handles multi modal outputs - ultra fast options will soon be available Their performance and price combination will almost become impossible to match
Vibe Code Your Own LLM You won't need a P.hD to create your own LLM and custom tune it to your use case anymore You can simply go from a text prompt to a custom-trained LLM in an hour. Simply ask the Abacus AI SuperComputer and use a GPU to train your custom…
🚨 Smaug-Nemotron is the best US open-source model in the agentic category - can run locally on your desktop! - tops agentic coding in US open-source - number #1 on cost/task - builds on Nvidia's latest model We are excited to push the boundaries of US open-s…
🚨 Kimi K3.1 Scores Below DeepSeek Flash - seems to be bench-maxxed on public benches - competitive on price per task - it's not Fable or Astra class - Deepseek is still king of open-source This is on LiveBench
Looking at public benchmarks is a total waste of time No lab worth its salt will release a model without checking and optimizing for benchmarks Literally every model is optimized for benchmarks and 3D renders that can go viral on X
The A.I. Daily