The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: https://t.co/bwn0sximKZ
Played around with Fable 5: a bike ride down Stockholm, Odenplan to Stadsbiblioteket.
One HTML file. Zero assets. Everything generated in code.
It has weak spatial intuition, It nails layout but has surprisingly shaky spatial understanding!
https://t.co/Fo1Vvix3y0
Claude Fable 5 is out. One-shotted a walkable 3D Gamla Stan (Stockholm) in Three.js, zero external assets, all textures procedural, runs straight in the browser.
Not perfect, but for one prompt? Wild.
Inspired by @c_nemri 's Sidi Bou Said.
https://t.co/vyDMynEH2e
For people who keep asking what to build in AI Engineering
> Build your own Reasoner (Chain of Thought implementation)
> Build your own Agent loop (ReAct pattern)
> Build your own Inference Server (in C++/Rust)
> Build your own Transformer from scratch (Attention is all you need)
> Build your own Vector Database (HNSW index)
> Build your own RAG pipeline
> Build your own Flash Attention kernel (CUDA)
> Build your own Quantization library (Int8/FP4 implementation)
> Build your own Mixture of Experts (MoE) routing layer
> Build your own Distributed training loop (FSDP/Tensor Parallelism)
> Build your own KV Cache paging system (like vLLM)
> Build your own Speculative Decoding system
> Build your own State Space Model (Mamba implementation)
> Build your own RLHF pipeline (PPO implementation)
> Build your own Small Language Model (SLM)
> Build your own Matrix Multiplication kernel
> Build your own LoRA (Low-Rank Adaptation) trainer
> Build your own Code interpreter sandbox
> Build your own DPO (Direct Preference Optimization) loss function
> Build your own Graph RAG system
> Build your own Model merger (Model Soups/Spherical Linear Interpolation)
> Build your own Interpretability tool (SAE - Sparse Autoencoders)
> Build your own Synthetic data generator
> Build your own Function Calling router
> Build your own Structured Output parser (Context Free Grammars)
> Build your own Multi-modal projector (CLIP implementation)
> Build your own LLM Eval harness
> Build your own Guardrails system (Input/Output filtering)
> Build your own Prompt caching mechanism
> Build your own Tokenizer (BPE implementation)
> Build your own Autograd engine (like Micrograd)
> Build your own Diffusion model (UNet + Scheduler)
> Build your own Vision Transformer (ViT)
> Build your own Whisper-style ASR model
> Build your own Text-to-Speech pipeline
> Build your own Semantic Router
> Build your own Knowledge Graph builder
> Build your own Data curation pipeline (MinHash/Deduplication)
> Build your own AI Gateway (Load balancing/Failover)
> Build your own Parameter Efficient Fine-Tuning (PEFT) library
> Build your own Text-to-SQL engine
> Build your own Recommendation system (Two-tower architecture)
> Build your own Embedding model
> Build your own Logit Processor
> Build your own Softmax kernel optimization
> Build your own Adversarial attack generator
> Build your own Audio Spectrogram transformer
> Build your own Neural Architecture Search
> Build your own Model Distillation pipeline
> Build your own Feature Store
> Build your own Database driver (for Vectors)
If a planner or logistics operator wants to know the impact of closing a road or building a new district, they have to spend weeks cleaning data, fighting with clunky desktop software, and exporting static charts. By the time the report was done, the city already moves on.
The future of urban planning isn't just "faster software." It’s a completely new way of interacting with spatial data.
An update to the interface for https://t.co/1nXZMGw7H6 to reflect this shift. Putting a Natural Language "Copilot" side-by-side with GPU-accelerated engine.
No need to manually build node networks anymore. Just ask a question in plain English: 🗣️ "Convert Sveavägen into a pedestrian-only zone. How does vehicle traffic reroute, and what is the spillover onto parallel streets?"
In seconds, the engine calculates 40,000 trips per second, reroutes the synthetic population, and gives you:
✅ The visual traffic cascade on the map.
✅ The exact number of affected trips (e.g., 132,000+ trips).
✅ The precise spillover bottlenecks (e.g., +14k trips on Birger Jarlsgatan).
Visit https://t.co/Ojdxkk8SfF to request access.
With Bonzai we are enabling dynamic traffic simulations and what was previously overnight batch jobs into interactive agent interactions. With our gpu accelerated routing algorithms we are 50x faster than osrm.
If you're interested in how Bonzai can help you send me a dm
"We can't simulate the city because we don't have perfect data."
Planners have been paralyzed by the "ground truth" problem. Legacy tools are so painfully slow that planners can only afford to run a simulation once. Because of that, the starting data has to be 100% perfect. And in a living, breathing city, perfect data doesn't exist.
But with the collision of GenAI and high-speed compute, the playbook has completely flipped. You no longer need perfect past data to command the future.
Here is how modern urban simulation can actually work:
1️⃣ Generate Synthetic Populations Instead of relying on outdated census or invasive tracking data, AI generates a "Synthetic Population." It creates millions of virtual residents whose behaviors mathematically mirror the real city, completely bypassing data privacy and availability bottlenecks.
2️⃣ Run simulations since we don't know exactly what the future holds, we don't run one simulation, let's run millions. A million different realities (a rainy Tuesday, a transit strike, a sudden population boom).
3️⃣ Deploy AI for Robust Optimization Instead of humans manually guessing the best road layout, AI agents test infrastructure changes against all million simulated realities. The engine finds the most resilient policy, the one that survives 99% of the chaotic futures we threw at it.
https://t.co/1nXZMGw7H6 is building a custom GPU-accelerated engine that routes 40,000 trips a second to make this exact workflow a reality. Not just predicting one fragile future; helping you optimize for all of them.
Early access: https://t.co/1nXZMGw7H6
The Gemini 3.1 demo proves the era of AI spatial intelligence is here.
At Bonzai, we are putting that predictive orchestration into the hands of city planners and logistics operators today.
We are currently in private beta in Stockholm.
DM me for early access! 🚀🏙️
https://t.co/Ojdxkk8SfF
We used Gemini 3.1 Pro to build a realistic city planner app. 🏙️
Watch how the model tackles complex terrain, maps out infrastructure, and simulates traffic to generate a high-quality visualization.
Congrats on the launch @simile_ai ! (and I am excited to be involved as a small angel.)
Simile is working on a really interesting, imo under-explored dimension of LLMs. Usually, the LLMs you talk to have a single, specific, crafted personality. But in principle, the native, primordial form of a pretrained LLM is that it is a simulation engine trained over the text of a highly diverse population of people on the internet. Why not lean into that statistical power: Why simulate one "person" when you could try to simulate a population? How do you build such a simulator? How do you manage its entropy? How faithful is it? How can it be useful? What emergent properties might arise of similes in loops?
Imo these are very interesting, promising and under-explored topics and the team here is great. All the best!
We're coooked I just used opus 4.6 to rewrite one of our routing algorithms in cuda . As long as you have a quantifiable metric and result to evaluate agent performance and a feedback loop you're set. The new routing algorithm in Cuda is 10x faster on a 3050 in prod on a a100 it will be 100x faster.