John

@garrytan Is it only 30k lines since it the lite version?

It's official. GStack for OpenClaw is here. When OpenClaw has to use Claude Code to do things (and it does this all the time) suddenly it can do it with wings. I created a special gstack-lite to keep OpenClaw tasks fast while making them think harder and get more done.

@skylar_b_payne @openclaw @jxnlco

unpopular @openclaw opinion I think gpt-5.4 with the right prompting can actually be better than opus 4.6. get ya pitchforks or join me in the promise land going all in on openai now that they got our boy @jxnlco

@Amank1412 OpenAI's burning $5 billion annually while Anthropic actually turns a profit in enterprise. Sometimes the tortoise beats the hare, especially when the hare's on fire

who's gonna win this AI race? > anthropic > openai

The ClawdBot mass exodus might actually be Anthropic's smartest business move yet. Here's what everyone's missing about those Claude cancellations... While developers rage about automated coding restrictions, Anthropic just solved their biggest problem: users who were essentially running industrial-scale operations on consumer pricing. Think about it...ClawdBot users were generating thousands of lines of code daily, consuming massive computational resources for $20/month. That's like running a data center on a home electricity plan. The math never worked. These power users were subsidized by casual subscribers asking Claude to help with emails and homework. Anthropic was essentially paying people to use their most expensive features. Now they're forcing a natural market segmentation: • Casual users stay on affordable plans • Heavy automation moves to enterprise pricing • Resource allocation becomes sustainable Sure, some developers are jumping ship to competitors. But those competitors will face the same economics eventually. You can't offer unlimited compute at consumer prices forever. The real winners? Anthropic's remaining users will likely see better performance as server loads normalize. Plus, this positions them for long-term profitability instead of venture capital life support. Sometimes the best business decision looks like the worst PR move 📊 What's your take - short-term pain for long-term gain, or did they just hand market share to OpenAI?

The pricing at $2.50 input means they've cracked inference costs that would make current enterprise deals look like highway robbery. That 2M context window isn't just an upgrade; it's basically saying "bring us your entire codebase and we'll understand it all" This will be huge if true. Open Ai's marketing and fundraising is next level too! Leak just enough specs to make competitors scramble while keeping the actual capabilities under wraps until launch.

GPT-6 rumored for April 14 from @iruletheworldmo Unverified leak but the details are worth watching through a hardware lens 2M token context window Natively multimodal $2.50/$12 per million tokens “Superapp” merging ChatGPT, Codex, and Atlas browser into one agent Now connect this to what just happened with @AnthropicAI @bcherny confirmed Anthropic is deprioritizing third-party tool access because agentic workloads don’t fit their capacity model. OpenClaw got cut from subscriptions. Token rationing is here Two of the biggest AI labs are telling you the same thing at the same time: Agentic compute demand is outrunning infrastructure Anthropic can’t efficiently serve it on TPUs. OpenAI is allegedly throwing every GPU at their next model and killing Sora to free up capacity. Both are rationing access This is the Co-Design Thesis The models are ready. The software is ready. The hardware isn’t keeping up. And the hardware that handles agentic workloads best (dynamic, unpredictable, CPU+GPU coherent, memory bandwidth dense) is NVIDIA 2M token context? That’s a Memory Wars problem. KV cache at that scale requires HBM4-class bandwidth. There is no workaround Every frontier lab is converging on the same bottleneck. And Jensen has been building for it for three years The race isn’t to build the best model anymore anon The race is to serve it

@JoelDeTeves Been running similar tests on 3090s; that IQ4_NL quantization really does punch above its weight class. The memory efficiency gains let you actually use these models instead of just loading them. Gemma4-26B-A4B is very fast, tool calling is lacking a bit though.

My 2 cents on Qwen3.5 vs Gemma4: Use Qwen3.5-27B if you have < 24 GB VRAM and need to fit a dense model, I like bartowski IQ4_NL - still my top choice for accuracy at this level, makes for a great agent too Use 9B if you have 16 GB or less For MOE / speed, Gemma4-26B-A4B is incredible. IMO it makes better decisions than Qwen3.5-35B-A3B and doesn’t loop. And it’s smaller! Great model for 24 GB cards, feels like the best balance of speed + performance for Hermes Agent Of course, some people may disagree - always DYOR that’s where the fun is 😎

@digitalshane_ Claude's basically admitting it writes spaghetti code then charges you to untangle its own mess 🍝 🤣 The irony? We're debugging AI with AI that can't debug itself.

I'm not joking, I had Claude write a big batch of code last night. I am troubleshooting rn. I asked it to review, It said this is trash code and it needs completely reworked. We are spending credits to run in circles.

@thdxr Amazon nearly went bankrupt betting everything on AWS infrastructure, now it just prints money.

engineers bias towards conservative approaches to business everyone loved when dario said he didn't want to overinvest because if he was wrong the whole business would die but you look at a lot successful companies and at some point they had to make the all or nothing bet

@Saboo_Shubham_ You’re kidding right?

OpenClaw is 🔥 and Peter is GOAT Send this to your OpenClaw Agent: models auth login --provider anthropic --method cli --set-default This is so much fun!

@OpenRouter @GoogleDeepMind You forgot the best part: $0.13/M input tokens $0.40/M output tokens

Gemma 4 from @GoogleDeepMind has hit 2.5B tokens so far: Two flavors: 31B dense, & 26B MoE (running 5x faster than the dense version now!) 256K context • native function calling • multimodal • configurable thinking • 140+ languages Try the 26B MoE: openrouter.ai/google/gemma-4…

@garrytan I personally never understood the love for OpenClaw. They seem to be polar opposites: you either love it or you hate it. I found it underwhelming and error prone.

Anthropic shutting down OpenClaw may turn out to be a strategic blunder, or strategic genius. The OpenClaw community will be the determiner of whether it is A or B. It's an interesting moment in history. Personally I never bet against open source.

@PawelHuryn I can't remember why I got this, but in testing it this morning, yes, it's good, but I'm sticking with Qwen. “Near AGI” it's a pretty funny take 🤪

Everyone's calling Gemma 4 "near AGI." Has anyone actually tried coding with it? I ran E4B on a 16GB MacBook via Ollama. Genuinely impressive model for its size. But it can't use tools in Claude Code. Chat works. Function calling doesn't. The model has its own tool call schema - and the tooling layer can't parse it yet. Benchmarks measure chat. Agents need tools.

BREAKING:🚨 NVIDIA just quantized Gemma 4 31B on Hugging Face 🔥 NVFP4 compression = 4x smaller weights with frontier-level accuracy. ✅99.7% of baseline on GPQA (75.46% vs 75.71%). 📈256K context window. 🧐Multimodal (text + images + video). vLLM-ready + Blackwell optimized. VRAM requirements: ⚡️Weights only: ~16–21 GB 🚀Everyday use: Runs on 24 GB GPUs 📈Full 256K context = 32 GB VRAM sweet spot (RTX 5090-class consumer GPUs) This is the 31B-class frontier model you can actually run locally on a high-end rig. Try it today👉 huggingface.co/nvidia/Gemma-4…

Google just cooked with Gemma 4. The new 31B model is already live on Ollama and holding its own against massive 397B-parameter models on the Arena leaderboard. For OpenClaw users, this is a perfect local agentic option. No API keys, full privacy, runs great on a single GPU. Just run: ollama pull gemma4:31b Local AI keeps getting stronger.

A 26B–31B model is now trading blows with Qwen 3.5’s 397B parameter monster on the Arena leaderboard. The 26B MoE version is especially insane: only a tiny fraction of parameters are active at any time, yet it’s still competing with models 12–15× its size. 🤯 I’ll be running some tests tomorrow.

@petergyang @OpenAI @openclaw Ollama Cloud: x.com/ollama/status/…

Codex team and @OpenAI have a huge opportunity right now to: 1. Tell @openclaw users how to switch to gpt subscription (I think it’s just telling the bot to switch the model?) 2. Fix GPT’s personality (maybe even sharing a prompt will help in the short term?). This is the main reason why ppl prefer using OpenClaw with Opus.

🦞Ollama's cloud is one of the best places to run OpenClaw. $20 plan is enough for most day to day OpenClaw usage with open models! To make the switch, all you need is to open the terminal and type: ollama launch openclaw Choose a model: kimi-k2.5:cloud glm-5:cloud minimax-m2.7:cloud If you are affected, Ollama welcomes you!! ❤️

@kenwheeler But, you are the one crying the loudest. Want to go to Brokeback with me and do some fishing?

imagine how much of a fucking cuck you have to be to show up in my replies simping for fucking anthropic

Announcing Textual Diff View! Add beautiful diffs to your terminal application. ⭐ Unified and split view ⭐ Line and character highlights ⭐ Many themes ⭐ Horizontal scrolling github.com/batrachianai/t…

@theo @willmcgugan Toad is safe

One piece of good news - T3 Code confirmed SAFE FOR CLAUDE SUBS 🫡 We FINALLY have explicit confirmation that tools wrapping Claude Code for local use are allowed