Manav Gupta

Won 2nd place in a crazzy hackathon by @Replit It was super fun. I built an virality engine by reverse engineering @AIatMeta's research paper on TRIBE v2 all in all a neuro hacking for viral reels. anyways, thanks to @amasad for giving free credits. and @networkschool for hosting it.

@benln posted a photo of cafe cursor in hanoi. a coffee shop in vietnam dedicated to cursor users. the AI dev scene in SEA is not emerging. it is already here. singapore, hanoi, bangalore, manila. the US just does not have the feed for it yet.

8/ DeepSeek was the shot. K2.6 is the confirmation. The next Chinese lab is already training. The question is not if. The question is what category gets hit next. My bet: coding agents. The benchmarks are already there.

7/ The real read: If you are building on closed models and praying for a price cut, you are gambling. If you are building on open models and praying for a capability bump, you just got it.

The DeepSeek playbook just ran again. Kimi K2.6 is the receipt 1/ January. DeepSeek drops a model, Nvidia loses $600B in a day, Altman makes ChatGPT free the same week. April. A different Chinese lab just did the same thing. Nobody is pricing it in yet. Kimi K2.6.

if your co-founder just said: "let us train our own model to save on API costs" they have not seen the Kimi K2.6 benchmarks yet. the answer is not training. the answer is switching providers every 90 days while the prices race to zero.

kimi k2.6 just did what deepseek did in january. SWE-Bench Pro 58.6. browsecomp 83.2. toolathlon 50.0. open-source. every closed model team just got less sleep tonight. #kimi #ai #opensource

Nous Research + Kimi Moonshot are running the Hermes Agent Hackathon. $25K pool. 16 days. Build an agent using Hermes 3 + the Kimi context API. If you want the exact arena most agent researchers will be watching this quarter, this is it. Link in comments. Tag your agent-building friend. #kimi #ai #nous

Q1 2026 venture: $297B globally. $242B into AI (81%). Four AI deals alone beat all of 2024's total venture: OpenAI $122B. Anthropic $30B. xAI $20B. Waymo $16B. If you're raising non-AI right now, you're raising on a 19% remainder. #deals #openAI #anthropic #xai #waymo

If you're an IC today: Stop benchmarking your own value against past-you. Benchmark against GPT-5.4 + 18 months. What's left after that is your actual moat. Follow @tensor___boy . I track this shift every day.

I run Cevi AI. Healthcare voice agents in production. I've been watching this frontier move every month. The gap between "AI can do your job" and "AI does your job" isn't intelligence anymore. It's deployment.

GPT-5.4 Hit 83% on GDPVal. What This Actually Means. GPT-5.4 scored 83% on GDPVal yesterday. If that number doesn't mean anything to you, it should. GDPVal measures how well AI does jobs with real economic value. Not benchmarks. Not riddles. Real work that pays humans. Let me break down what this actually means.