MarkTechPost

Signals

Latest developments

Anthropic Launches Claude Code Projects in Beta: Parallel Cloud Sessions That Keep Running After You Close Your Laptop MarkTechPost · 13 hours ago · 1 OpenAI Releases a Model Misalignment Disclosure Framework With 3 Review Tracks and 6 Incident Reports From RL Training MarkTechPost · 1 day ago · 30 Google Research Introduces Retrieve-for-Train (R4T): An RL-Compiled Diffusion Retriever for 12× to 20× Faster Query Fan-Out MarkTechPost · 1 day ago · 34 Nunchux AI Introduces VC-Attention: A Training-Free Low-Bit Attention Kernel That Speeds Up Video Diffusion Transformers MarkTechPost · 1 day ago · 5 Stanford Researchers Release Paper2Agent: Turning Research Papers Into AI Agents That Reproduce Results and Run on New Data MarkTechPost · 1 day ago · 50 Knowledgator Releases GLiFormer: A 575M-Parameter Encoder That Hits 91.10 F1 on Nested JSON Extraction Without Generating Tokens MarkTechPost · 1 day ago · 39 Prior Labs Releases TabPFN-3.5: A Tabular Foundation Model That Beats the Winning Otto Kaggle Solution With Default Settings MarkTechPost · 2 days ago · 12 Nums AI Releases Causilo: A Tabular Foundation Model That Tops TabArena Among Single Models MarkTechPost · 2 days ago · 31

Q3 2026

◆ Anthropic relaunches Claude Code Projects in beta to coordinate multiple parallel cloud sessions with shared memory, goals, and files Feature update

◆ Stanford researchers released Paper2Agent, a system that converts research papers and codebases into MCP server–backed AI agents for reproducible execution on new data Research publication

◆ OpenAI released a framework for tracking, investigating, and disclosing model misalignment incidents and published six additional concerning behavior reports Other

◆ Google Research introduces Retrieve-for-Train (R4T), a training-time RL-compiled approach for faster set-valued AI retrieval Research publication

◆ Articles describe production harness mechanisms to prevent long-horizon LLM agent failures, including context management and execution loop limits Other

◆ Researchers released a “Fly Language Model” that uses a fruit-fly connectome as a reservoir on top of a frozen language model backbone and trained only a small readout adapter Research publication

◆ Cohere released North Small Translate, an open-weight mixture-of-experts machine translation model available via API and with non-commercial self-hosting checkpoints Model release

◆ Sakana AI releases Fugu Max and Fugu Ultra v2 multi-agent orchestration models via a hosted, OpenAI-compatible API Model release

◆ AWS released an open-source Pizza Bot desktop app that presents background AI agent outputs and approvals in an email-style inbox interface Open source release

◆ DeepSeek released the multimodal Mixture-of-Experts model DeepSeek-V4.1-Flash, featuring a 1M-token context window and KV-cache efficiency improvements Model release

◆ OpenAI introduced the Agents API, a public-beta managed API for building and running long-running cloud-based AI agents Feature update

◆ Google DeepMind releases AlphaGenome Atlas, a public resource with precomputed AI predictions for the effects of 9 billion single-nucleotide human DNA variants Open source release

◆ MBZUAI IFM releases K2 Horizon, a suite of six Apache 2.0 open foundation models with training artifacts and deployment support Open source release

◆ OpenAI launches GPT-6 Astra, a computer-use model with a 1.05M-token context window and staged enterprise/API access Model release

◆ Meta released Muse Spark 1.3, an agentic coding model with efficiency improvements and new availability via Muse Code and the Meta Model API Model release

◆ Google DeepMind and Google Research introduced WeatherNext 3, an updated AI model for global weather forecasting that refreshes hourly using real-time satellite observations Model release

◆ Google DeepMind launches Gemini 3.8 Flash and Gemini 3.8 Flash Cyber and introduces the Fairwind program for early access to the cyber-focused model Model release

◆ Perplexity launched Hybrid Compute on the Mac, enabling AI agents to split tasks between cloud models and on-device models for privacy-aware processing Feature update

◆ Google launches Gemini 3.8 Flash and related agentic video understanding capabilities for its Gemini Flash models Model release

◆ Anthropic announced Enterprise Frontier Safeguards (EFS) to add customer-controlled zero data retention and automated misuse detection for its frontier models Feature update

◆ Google Research releases TimesFM-3, a zero-shot multivariate time-series forecasting foundation model Model release

• 17 Sep Anthropic Launches Claude Code Projects in Beta: Parallel Cloud Sessions That Keep Running After You Close Your Laptop
• 17 Sep OpenAI Releases a Model Misalignment Disclosure Framework With 3 Review Tracks and 6 Incident Reports From RL Training
• 17 Sep Google Research Introduces Retrieve-for-Train (R4T): An RL-Compiled Diffusion Retriever for 12× to 20× Faster Query Fan-Out
• 17 Sep Nunchux AI Introduces VC-Attention: A Training-Free Low-Bit Attention Kernel That Speeds Up Video Diffusion Transformers
• 16 Sep Stanford Researchers Release Paper2Agent: Turning Research Papers Into AI Agents That Reproduce Results and Run on New Data
• 16 Sep Knowledgator Releases GLiFormer: A 575M-Parameter Encoder That Hits 91.10 F1 on Nested JSON Extraction Without Generating Tokens
• 16 Sep Prior Labs Releases TabPFN-3.5: A Tabular Foundation Model That Beats the Winning Otto Kaggle Solution With Default Settings
• 16 Sep Nums AI Releases Causilo: A Tabular Foundation Model That Tops TabArena Among Single Models
• 15 Sep Inside NVIDIA’s cuDNN Graph API: Fusion, Autotuning, and Plan Reuse with cuDNN Frontend
• 14 Sep Agent Harness vs Agent Framework vs MCP: Which Layer Owns the Loop, State, Tools, Permissions, and Recovery
• 14 Sep Reward AI Releases OM-1: A Robot Policy Trained on Human Demonstrations Only, With No Teleoperation or On-Robot Data
• 14 Sep Sakana AI Researchers Introduce PC-ALM, a Layer-Local Alternative to Backpropagation That Trains 1000-Layer Networks
• 13 Sep Hierarchical NeRF with JAX3D for Volumetric Rendering, Novel-View Synthesis, and 3D Reconstruction
• 13 Sep A Princeton Researcher Proposes Recurrent Looped Transformer (RLT) that Carries Decoder State across Every Token, Fixing 96 Blocks per Token with Unbounded Temporal Depth
• 13 Sep AWS Introduces Pizza Bot: An Open Source Inbox for Background AI Agents
• 13 Sep Context Engineering Inside the Harness: 4 Mechanisms That Beat Context Overflow and Goal Loss on Long-Horizon Tasks
• 12 Sep Fly Language Model (FLM) Wires the Full Fruit Fly Connectome Into a Frozen 1.2B LLM, and Its Own Controls Show the Wiring Does Not Help
• 11 Sep Can LLMs Engineer Their Own Agent Harness? ByteDance Seed’s HarnessDev Says Only 34 of 64 Changes Generalize
• 11 Sep Anthropic Adds Plugin Evals to Claude Code: 6 Grader Types, a No-Plugin Baseline, and a CI Gate for Skills
• 11 Sep Cohere Releases North Small Translate: A 218B MoE Translation Model That Scores 83.6 on WMT26 Across 50 Languages
• 11 Sep Sakana AI Launches Fugu Max and Fugu Ultra v2 for Cheaper, Stronger Multi-Agent Orchestration
• 10 Sep Meet Redis LangCache: A Managed Semantic Cache That Cuts LLM API Costs by Up to 90% and Returns Cache Hits Up to 15x Faster
• 10 Sep NVIDIA Details BioNeMo Inference Runtime (BioIR): 2.90x Higher Boltz-2 Folding Throughput and 58.5K Residues per GPU-Hour on 8xH100
• 10 Sep OpenAI Launches the Agents API in Public Beta, Putting the Codex Harness Behind One API Call
• 10 Sep DeepSeek AI Released DeepSeek-V4.1-Flash with 1M Context, FP4 KV Cache, and Cross-Layer Attention Reuse
• 10 Sep LandingAI Releases Agentic Document Extraction Gen2 with DPT-3 Pro and DPT-3 Verity
• 9 Sep Google Open-Sources Mantis: A Modular Skills Toolkit That Lets Coding Agents Find, Reproduce and Patch Vulnerabilities
• 9 Sep Gradium Launches Voice Design: Write a Prompt, Get a Brand New Synthetic Voice in Seconds
• 8 Sep Google DeepMind Releases AlphaGenome Atlas With Precomputed Molecular Effect Predictions and AVI Scores for 9 Billion Human DNA Variants
• 8 Sep Reducto Releases r-1: A Single Pass Document Parsing Model That Cuts Errors 20% at 1 Cent Per Page
• 7 Sep Axis Robotics Releases AXIS: A Browser-Based Data Engine With 207 Robot Manipulation Tasks and 50,129 Trajectories
• 7 Sep IFM Releases K2 Horizon: Six Apache 2.0 Models From 0.9B to 375B
• 6 Sep H Company Releases NeoMME: A Family of 260M and 800M Single-Tower Multimodal Encoders That Drop the Vision Tower and Causal Decoder
• 6 Sep UC Berkeley Researchers Release CUA-Lite, an Open Platform Unifying Sandboxes, Data, Evaluation and RL for Computer-Use Agents
• 6 Sep Perplexity Details Its GPU Embedding Stack: How Ivy, Tulip and ROSE Serve pplx-embed
• 5 Sep Adaption Labs Introduces ‘Invent a Dataset’: Training Data Generated From a Task Description, Not a Seed Corpus
• 5 Sep Google Launches Agentic Video Understanding for Gemini Flash Models, Cutting Video Tokens by Up to 88%
• 4 Sep Google DeepMind’s WeatherNext 3 Trains on Weather Station Observations to Deliver 5 km Global Forecasts, Refreshed Every Hour
• 3 Sep OpenAI Releases GPT-6 Astra: A 1.05M-Context Computer-Use Model Gated Behind a ‘Critical’ Cyber Threshold
• 3 Sep Anthropic Released Claude Commerce Agents: An Apache-2.0 Blueprint for Shopping and Merchant Agents Across Retail, Travel, Telecom and Entertainment
• 3 Sep Perplexity Open Sources Lily: A Rust + Metal Inference Engine for Qwen3.6-35B-A3B on Apple Silicon
• 2 Sep Qwen Developers Open-Sources zg (zvec-grep): A Local-First Search Layer Unifying ripgrep, BM25, and Vector Search
• 2 Sep Meet Switchyard: A Rust Proxy and Library That Routes and Translates LLM Traffic Across OpenAI and Anthropic APIs
• 2 Sep Google DeepMind Releases Gemini 3.8 Flash and Gemini 3.8 Flash Cyber: One Core Model, Two Access Envelopes
• 2 Sep Anthropic Introduces Enterprise Frontier Safeguards (EFS): Zero-Data-Retention Privacy Plus Cross-Session Misuse Detection
• 2 Sep Meta Superintelligence Labs Releases Muse Voice Transcribe: One Real-Time Model for Streaming ASR, Diarization, and Endpointing
• 2 Sep Perplexity Releases Hybrid Compute on Mac: Cloud Agents Orchestrate Down to a Local Model, Gated On Device
• 1 Sep Gradium AI Releases New Default TTS Model: 81.0% Hard-Case Pass Rate at 216 ms Time-to-First-Audio
• 31 Aug Keenable AI Open-Sources NEEDLE: A Live Search Benchmark That Rebuilds Its Query Set Every Hour
• 31 Aug Google AI Releases TimesFM-3: A 330M Parameter Zero-Shot Foundation Model For Multivariate Time Series Forecasting

Relationships

Products & technology

• Develops NVIDIA · 1 source
• Derived from Claude Code Projects · 1 source
• Integrated with GlucoFM · 1 source

Partnerships

• Partnered with TwIL-LM · 1 source
• Partnered with HarnessDev · 1 source
• Partnered with Anthropic · 1 source
• Partnered with NeoMME · 1 source
• Partnered with DeepSeek Harness v0.1 · 1 source
• Perplexity partnered with this company · 1 source
• Enterprise Frontier Safeguards (EFS) partnered with this company · 1 source
• CoreWeave partnered with this company · 1 source
• Nums AI partnered with this company · 1 source