AI Agents
CLAUDE OPUS 5 LANDS ACROSS THE STACK
The newest Anthropic model is now live in langchain, Cline, and llama-index, with native support for extended reasoning and 1M context windows.
Read the wire →
Topic · AI / ML
AI / ML news, every morning.
Daily updates from the AI and machine-learning ecosystem - PyTorch, Hugging Face, the Anthropic and OpenAI SDKs, local LLMs, and the agent frameworks.
Last 7 days
Local LLMs
OLLAMA LANDS LAGUNA SUPPORT AND CRUSHES MEMORY LEAKS WHILE SGLANG HITS V0.5.16 WITH CONFIDENCE-DRIVEN SPECULATIVE DECODING
Ollama shipped three critical performance and reliability fixes for Metal residency and concurrent access patterns, while SGL-Lang released 0.5.16 with a new speculative algorithm hitting 383.7 tok/s on DeepSeek-V4.
Read the wire →
Hugging Face
TRANSFORMERS OVERHAULS LINEAR ATTENTION WHILE DEPRECATING LEGACY RESPONSE SCHEMA
The transformers library is retiring its fragile response_schema prototype in favor of streaming-compatible parsing, while simultaneously refactoring every linear attention model to use standardized convolution patterns.
Read the wire →
PyTorch
CRITICAL OIDC INJECTION IN DOCS PREVIEW WORKFLOW PATCHED
PyTorch's docs-preview CI trusted fork-controlled artifacts in a context with token-write permissions, exposing the entire build pipeline to code injection.
Read the wire →
JAX FIXES REMAT3 CONSTANT HANDLING CRASH, GOOGLE CLOUD PYTHON CENTRALIZES REST TRANSCODING
Matthew Johnson patched a critical remat3 regression where closed-over constants were silently dropped, causing immediate crashes in production rematerialization pipelines [ref:1].
Read the wire →
Meta
REACT FIBER FIXES SINGLETON PROPERTY BLEED, REACT NATIVE STRIPS DEAD CODE
React plugged a critical Fiber bug where singleton cleanup was nuking third-party attributes, while React Native dropped unused timing code from iOS binaries.
Read the wire →
OpenAI
CODEX HARDENS MCP RUNTIME STABILITY ACROSS FIVE CRITICAL FIXES
OpenAI shipped a coordinated hardening of the Model Context Protocol layer across Codex, patching race conditions in thread startup, resource access, and authority management that could leave stale configurations in production.
Read the wire →
Anthropic
CLAUDE AGENT SDK FIXES STDIN CLOSURE BUG IN BACKGROUND TASKS
The Python SDK's query method was prematurely closing stdin when background subagents were still in flight, breaking tool calls and silently bypassing hooks - now fixed in v0.2.127.
Read the wire →
AI Agents
LANGCHAIN SHIPS COORDINATED PROVIDER UPDATES WITH LANGSMITH GATEWAY SUPPORT
All four core LangChain packages bumped to 1.5.1 overnight with unified LangSmith gateway routing and provider-specific fixes across OpenAI, Anthropic, and Fireworks.
Read the wire →
Local LLMs
LLAMA.CPP SHIPS FIVE RELEASES IN 24 HOURS, FIXES SPECULATIVE DECODING ACROSS THE STACK
The inference stack is consolidating around speculative decoding, and the bugs surfacing there are getting squashed across three major projects simultaneously.
Read the wire →
Hugging Face
TRANSFORMERS SHIPS FSDP DISTRIBUTED TRAINING STACK, HUB LIBRARY PLUGS REDOS HOLE
Hugging Face landed distributed training orchestration in transformers while plugging a regex vulnerability that could stall untrusted card parsing for minutes.
Read the wire →
PyTorch
PYTORCH AUTOGRAD GETS 7% FASTER, AOTI FIXES SILENT FAILURES
Interned attribute names in autograd.Function shaved microseconds off the hot path while AOTI's scatter operations now properly report errors instead of silently corrupting results.
Read the wire →
JAX TIGHTENS XLA OPTIMIZATION CONTROLS, GOOGLE CLOUD PYTHON SETS UP FEATURE GATES
JAX makes XLA optimization levels human-readable while Google Cloud Python client libraries prepare a unified feature gating system for observability.
Read the wire →
Meta
REACT DEVTOOLS BRIDGE HARDENS AGAINST RECONNECTS AND CORRUPTION
React's DevTools layer just got bulletproof: hoxyq landed four interconnected patches that eliminate silent state corruption, validate invariants before mutation, and handle extension reconnects with message buffering and handshakes.
Read the wire →
OpenAI
SDK RELEASES LOCK IN PROMPT CACHING AND SPEND CONTROLS
Both Node and Python clients shipped v6.49.0 and v2.48.0 overnight with admin spend limit APIs and relaxed prompt cache validation.
Read the wire →
Anthropic
BUFFA v0.9.1 PATCHES REGRESSION THAT BROKE ORDINARY SCHEMAS
Buffa shipped an emergency patch overnight after v0.9.0 broke codegen on schemas with a few hundred proto files, small enough to hit in production.
Read the wire →
AI Agents
LANGCHAIN REFRESHES MODEL PROFILES, OPENROUTER 0.2.7 SHIPS
OpenRouter integration lands with updated model profiles including Google's Gemini 3.5 Flash Lite, while core token counting gets a performance fix for multi-tool agents.
Read the wire →
Local LLMs
OLLAMA AGENTS NOW IMPORT SKILLS FROM CODEX AND CLAUDE
Ollama shipped a skills import system that lets coding agents pull validated skill directories from external sources, while simultaneous fixes to GLM tool-call parsing and Laguna model alignment prevent agents from silently failing.
Read the wire →
Hugging Face
TRANSFORMERS HARDENS AGAINST PYTORCH FRAGMENTATION WHILE TRL SIMPLIFIES DISTILLATION
Transformers plugged a cascading import failure that breaks downstream CI on older PyTorch versions, while TRL rips out dead code to lock DistillationTrainer into prompt-only datasets.
Read the wire →
GENAI AUDIO TRANSCRIPTION SHIPS, PROTO-PLUS FIXES RACE CONDITIONS
Google's AI SDK gains native audio processing while the core Python cloud libraries ship critical stability fixes across the stack.
Read the wire →
Meta
REACT SERVER DOM TURBOPACK GETS DYNAMIC CHUNK MERGING
React's Turbopack integration now supports an experimental array format that lets you dynamically merge or unmerge chunks based on what's already loaded in the browser.
Read the wire →
OpenAI
CODEX CONSOLIDATES THREAD STARTUP, NODE SDK ADDS REALTIME AND STANDARD SCHEMA SUPPORT
Codex refactored its threading and database layer for consistency while the Node SDK ships four new schema helpers and fixes audio streaming in a single release push.
Read the wire →
Anthropic
CLAUDE FOUNDATION MODELS ADDS APP ATTEST, PLUGIN ECOSYSTEM STABILIZES
iOS developers can now ship Claude calls without API keys in the binary, backed by Apple's Secure Enclave.
Read the wire →
Hugging Face
DATASET VIEWER LOCKS DOWN ARROW, FUNES SHIPS GROUNDED ASK
Hugging Face security teams moved overnight to contain a critical Arrow IPC parsing vulnerability in dataset-viewer while shipping three production hardening releases across Repo2RLEnv, optimum-executorch, and funes.
Read the wire →
JAX CUTS ASYNC COLLECTIVE PRIMITIVES, GENAI SDK ADDS LIVE API VOCABULARY
JAX is simplifying its async collective operations while the Python GenAI SDK ships custom vocabulary support for the Live API, marking a shift toward leaner abstractions and richer real-time capabilities.
Read the wire →
Meta
React 19 Cuts Server Component Decoding Time, React Native Fixes requestIdleCallback Regression
React shipped performance wins across 19.0.8 and 19.1.9 while React Native patched a critical timeout bug that broke requestIdleCallback tasks.
Read the wire →
OpenAI
CODEX FIXES WINDOWS EXECUTION BUGS WHILE TIGHTENING SECURITY REVIEW GATES
Codex shipped critical fixes for Windows process management and terminal navigation that were silently breaking execution sessions on Windows machines.
Read the wire →
Anthropic
BUFFA FIXES V0.9.0 REGRESSION, CLAUDE PLUGINS ECOSYSTEM GAINS 9 NEW INTEGRATIONS
Buffa's codegen was silently rejecting moderately large schemas due to a memory bound meant only for untrusted input, but a fix shipped overnight restores throughput and clears the path for googleapis-scale proto sets.
Read the wire →
Hugging Face
TRANSFORMERS FIXES RECURRENT MASKING BUG; MOON-IDE KILLS PAIRING WINDOW
A critical padding mask bug in transformers is now fixed for chunked prefill and cache continuation, while moon-ide removes dead startup ceremony and hardens the remote bridge client.
Read the wire →
PyTorch
DYNAMO REVERTS BREAKING CHANGE, EXECUTORCH CLEANS UP DEPRECATED TYPES
PyTorch reverted a Dynamo optimization that broke internal tests, while ExecutorTorch is aggressively deprecating c10 shims in favor of standard library types.
Read the wire →
JAX SHIPS EINSUM PRIMITIVE AS FOUNDATIONS FOR DISTRIBUTED COMPUTING OVERHAUL
JAX's hijax layer now supports einsum as a first-class primitive, setting up the infrastructure for major distributed execution improvements across the stack.
Read the wire →
Meta
REACT PATCHES HIDDEN COMPONENT STATE BUG, REACT-NATIVE FIXES KEYBOARD TRAP
React shipped a critical fix for useSyncExternalStore mutations being missed when subtrees toggle visibility, while React-Native plugged an Android keyboard that won't close on unmount.
Read the wire →
OpenAI
CODEX TIGHTENS MEMORY HANDLING AND AUDIO PRESERVATION ACROSS THREADS
Codex shipped five structural fixes overnight that close gaps in memory reconciliation, audio attachment handling, and Windows permission checks that could silently corrupt state in production.
Read the wire →
Anthropic
RECEIPTS PLUGIN GETS USER-FOCUSED OVERHAUL, 10 PLUGINS QUEUE FOR VALIDATION REVIEW
The receipts plugin shipped a docs refresh that strips away defensive explanations and leads with what users actually need, while the plugin fleet cycles through validation on both the official and community sides.
Read the wire →
Hugging Face
TAU FIXES SESSION CORRUPTION BUG, SERGE ADDS GPU VERIFICATION LOOP
Tau's Unicode line separator bug was silently bricking user sessions. The fix is retroactive and ships today.
Read the wire →
PyTorch
PYTORCH SHIPS BUILD FIX WHILE HELION TUNES H100 KERNELS TO DEFAULT
A critical build regression in cusparselt.cpp is now patched, while the kernel autotuner promotes its pointwise seed heuristic to production defaults on H100 and B200.
Read the wire →
JAX TIGHTENS PPERMUTE VALIDATION TO CATCH SILENT CORRECTNESS BUGS
JAX is shipping a correctness fix that will break code relying on undefined behavior, and that's exactly the point.
Read the wire →
Meta
REACT FIZZ UNLOCKS NESTED ANIMATIONS FOR STREAMING SERVERS
React's server renderer now supports nested View Transition animations during Suspense reveals, closing a gap that forced SSR apps to choose between progressive enhancement and smooth animations.
Read the wire →
OpenAI
CODEX FIXES RACE CONDITIONS AND STABILIZES STREAMING PIPELINE
Codex shipped five architectural fixes overnight that eliminate liveness races in side conversations, stabilize the Markdown streaming stack, and cut TUI layout time significantly.
Read the wire →
Anthropic
PLUGIN VALIDATION SWEEP ACROSS ALL THREE REPOS
Eight plugin dependencies shipped validated updates across the official and community catalogs overnight, with mattpocock-skills joining the community collection.
Read the wire →
Hugging Face
MLCLAW TIGHTENS AUTHENTICATION, TAU SHIPS REASONING PERSISTENCE
MLClaw fixed a critical authentication gap where private Space edge cookies weren't reaching approval requests, forcing a shift to protected Space defaults across all hosted gateways.
Read the wire →
PyTorch
HELION BACKEND REFACTOR UNLOCKS MULTI-DEVICE COMPILATION; DYNAMO FIXES FORWARD-AD TANGENT BUG
Helion shipped a four-part backend architecture overhaul that finally separates printer, lowering, and backend registration logic by device, eliminating the need to edit 17 core files for every new backend.
Read the wire →
JAX BACKWARD PASS NOW LOGS DEBUG DATA. CLOUD AUTH TIGHTENS mTLS PRECEDENCE
JAX's autodiff pipeline shipped structured logging from transpose rules, letting you inspect what's happening inside the backward pass without tanking performance.
Read the wire →
Meta
REACT NATIVE TRANSLATION PIPELINE ROLLING BATCH 1
The automated translation update for React Native shipped its first batch across messaging and composer modules overnight, processing multiple Android resource files.
Read the wire →
OpenAI
CODEX HOTFIX CORRECTS GPT-5.6 CONTEXT WINDOWS AFTER BUNDLED INSTRUCTION REFRESH
Codex 0.144.6 shipped overnight to fix context window mismatches across GPT-5.6 Sol, Terra, and Luna after bundled instruction metadata got out of sync with the models themselves.
Read the wire →
Anthropic
CLAUDE CODE 2.1.215 SHIPS WITH SKILL BEHAVIOR CHANGES
Claude Code no longer auto-invokes verify and code-review skills - you now call them explicitly when needed.
Read the wire →