Mistral Studio just rolled out a major upgrade, transforming the platform into an end-to-end hub for building and hosting AI applications. Developers can now wire up reusable knowledge connectors, deploy reasoning agents and deterministic workflows with a single command, and turn production run traces into evaluations to continuously refine their apps.
Anthropic is working on a collaborative mode for Claude Code Projects. The platform's Projects page now summarizes recent thread activity, while Anthropic's Lydia Hallie teased yesterday that multiplayer functionality is in the works. Why it matters: Adding real-time team collaboration could make shared AI coding workflows far smoother.
Lattice just launched a local LaTeX workspace built for researchers, packing writing, PDF previews, and citations into one desktop hub. It features two-way sync with Overleaf and connects directly to coding agents like Claude Code to fix build errors and refine papers, cutting out constant app switching.
Anthropic just upgraded Claude Code cloud sessions for Pro, Max, and Ultra subscribers. The AI can now automatically figure out project dependencies, network access, and required API keys to spin up and run code in the cloud with minimal manual setup.
Anthropic just shared a look into how AI is transforming scientific research. Instead of researchers manually writing and weighing competing theories, AI can now write, run, fit, and tweak models as code to rapidly uncover the explanations backed by actual data.
Meta just rolled out Spaces inside its Muse app, letting users spin up personalized interactive trackers and boards simply by describing what they need in chat. The spaces pull data from connected apps to update in the background and can be pinned or scheduled across your tabs. Collaborative sharing is coming next, backed by specialized models trained on Muse Wonder.
A new report from safety nonprofit METR estimates Anthropic’s model R&D is moving nearly 2x faster, with AI driving about 80% of the jump. The evaluation also teased unreleased internal models like Model 4, suggesting the recursive AI self-improvement loop is already quietly running inside the lab.
Serval is recruiting engineers to build Catalyst, an AI agent that automates IT technician workflows from natural language prompts. Teams can test and approve the agent-generated code before scheduled execution. To lure talent, the startup is offering every new engineering hire a weeklong, $10,000 paid vacation before their first day.
Yukon proved capable of keeping table latency under two minutes during a one-hour, 1M events-per-second benchmark run. It saves tables in open formats directly to your cloud storage and logs ongoing changes, allowing AI agents to inspect what updated without scanning whole tables.
Thinking Machines just unveiled Inkling 2 Small, a 594B-parameter MoE model with only 17B active parameters built for autonomous agentic coding and visual reasoning. It can run experiments, fix bugs, and refine interfaces using visual feedback in the loop. Full weights are slated to drop under Apache 2.0 next week, but you can already test it on Tinker Playground.
Oumi just launched an agent-building platform powered by FAiE, an AI engineer designed to handle everything from training to verification. Users can simply correct mistakes in chat, prompting FAiE to update datasets, retrain the model, and measure gains in roughly 15 minutes instead of months of engineering.
Coding agent platform Conductor just joined Meta to build what its team calls the next interface for developers. Working alongside MSL, they plan to launch something new early next year. Why it matters: Meta appears to be assembling the pieces for its own AI coding super app.
ChatAndBuild just launched Living Brain, a persistent memory layer that lets ChatGPT, Claude, Cursor, and custom agents read and edit the same context. Users can switch between AI apps without re-explaining background details. It boasts a 464ms median retrieval latency and up to 22.6x fewer tokens per query.
Nace AI just launched a 120% cashback promotion for teams switching to Drex Max, its decision model for AI agents. Developers can migrate by changing three environment variables and earn $1.20 for every dollar spent. Nace claims 1,000 browser tasks on MiniWoB++ cost $2.80 with Drex Max, compared to $31 with Claude Code and $25 with Codex.
OpenAI just announced it will acquire Pierre Computer Company to upgrade Git workflows and code review inside Codex. Pierre's Trees and Diffs libraries already power file navigation and change reviews in the Codex app, and its open-source tools will remain open. Why it matters: as software teams deploy multiple coding agents in parallel, OpenAI wants to redefine modern version control for the agentic era.
Ricursive Intelligence just released Ripple-65B Preview, a dense open-weight model with a 1M context window trained for $32M. Skipping MoE complexity, it scored 80.52 on TerminalBench 2.1, beating massive frontier open-weight models over ten times its size in agentic coding.
Anthropic just open-sourced FlashProve, a provable inference system using cryptographic zero-knowledge proofs to verify an AI response actually came from the claimed model. Verifiers do not need weights or GPUs, checking proofs on a standard CPU in seconds, though generating proofs still takes 72 to 94 times more GPU compute.
Sandflow AI just open-sourced sand3d, a Python library built on Blender that lets LLMs construct parametric 3D objects and robotic joints in readable code. The entire API fits in a 3,600-token prompt—using up to 6x fewer tokens than raw Blender script—and launches alongside a massive 100,000-object open dataset.
Demos circulating on X show an unreleased Claude checkpoint dubbed "Fable 5.5" generating animations without any video or image models. In one test, it autonomously wrote Python code for over two hours to produce a standalone HTML animated film. Anthropic has not confirmed the model exists.
DeepPrinciple just dropped Hydrogen-1.0, an open-weight, 753-billion-parameter model designed for chemistry, materials science, and biology. Fine-tuned on real laboratory workflows, it can write and self-debug simulation code, decipher NMR spectra, and compile protocols directly for lab automation. It even matches or outperforms GPT-6-Astra on several core scientific benchmarks.
