Deep Learning Weekly
Newsletter / Aggregator
Newsletter aggregating deep learning and AI research/industry news.
Recent activity
-
Anthropic files S-1 with the SEC for proposed initial public offering, The sleep framework: A new way for LLMs to retain knowledge in continual learning, and many more!
Read more → -
Researchers have found that decision-model controllers, which delegate small decisions to cheaper models, raise agent cost without improving success, according to a study on Muse Glimmer 30B, an on-device agentic model trained on over 100 languages. A new open-source project, OpenGCUA, aims to develop and open the full stack for generalist computer-use agents, which can complete tasks on users' computers through visual and programmatic interfaces. A study on open-weight AI models found that they are 6-7 months behind on cyber security, with data analysis by Z.ai's GLM-5.3 model. AI summary
Read more → -
Claude Haiku 5.5 has been released with a 0.10/0.50 per million tokens pricing, achieving 39.2% on Terminal-Bench 4.0 at max effort against GPT-6 Luna's 16.4%. Meta's Muse Wonder and Muse Wonder Full models have achieved frontier performance, but the weaker one is available via API, scoring 52 at 4/20 per million tokens. A new scaffold called SELFCOMPACT has been proposed for self-compacting language model agents, which allows the model to decide when and how to compact, resulting in improved performance and reduced token cost. AI summary
Read more → -
Gemini 4 Argon, Open and Emergent Problems in Agentic Privacy and Security, a paper on Rethinking Cross-Tokenizer On-Policy Distillation, and many more!
Read more → -
Introducing GPT-6 Sol and Luna, AI Evals: Everything You Need to Know, a paper on RRSI: Regularized Recursive Self-Improvement of Agent Harnesses, and many more!
Read more → -
Introducing GPT-6 Sol and Luna, AI Evals: Everything You Need to Know, a paper on RRSI: Regularized Recursive Self-Improvement of Agent Harnesses, and many more!
Read more → -
We Must Pace the Frontier, ToolGrad: Efficient tool-use dataset generation with textual "gradients", a paper on The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement, and many more!
Read more → -
Introducing ChatGPT Images 2.5, Linguistic drift at the frontier, a paper on Random Attention: Rethinking KV Cache Eviction for Efficient Reasoning, and many more!
Read more → -
Introducing Claude Fable 5.1 and Claude Mythos 5.1, Evaluating LLMs Under Production Parity, a paper on Automated Researchers Can Reliably Mitigate Alignment Failures, and many more!
Read more → -
GLM-5.3-Flash, Evaluating Explanations of LLM Behavior in the Wild with Counterfactual Experiments, a paper on FreeToken: Efficient Edge-Native MoE Serving with Bandwidth-Adaptive Execution, and many
Read more → -
FLUX Upscale: 2K and 4K for Video, Empty shelves or lost keys? Recall is the bottleneck for parametric factuality, a paper on VibeWorlding: Can Multimodal Agents Construct 3D Open Worlds End-to-End?,
Read more → -
Introducing Grok 4.6, TutorMoments: Do AI tutors know when to help and when to hold back?, a paper on BDH-CQ: In-Context Learning with Recurrent Latent Reasoning, and many more!
Read more → -
Qwen3.8-Max, A verifiable autonomous research framework via Chain-of-Evidence, a paper on From RLVR to RLSVR: Task Transformation Induces Self-Verifiable Rewards for Open-Ended LLM Self-Improvement, a
Read more → -
Claude Opus 5 is highlighted alongside Digibee’s use of Opik to construct prompts that power their AI-native integration platform, and a new paper introduces Progress Reward Modeling for improving robotic learning. The issue also mentions additional developments and resources in deep learning. AI summary
Read more → -
Kimi K3, Beyond the Single Trace: How We Built Agent Diagnostics for Opik, a paper on Belief Dynamics Reveal the Dual Nature of In-Context Learning and Activation Steering, and many more!
Read more → -
Thinking Machines' Inkling, How We Optimized Opik’s MCP Server for Cost & Performance, Metacognition in LLMs: Foundations, Progress, and Opportunities, and many more!
Read more → -
Introducing Grok 4.5, How Internal Optimizations Led to Comet Cost Intelligence, a paper on AlayaWorld: Long-Horizon and Playable Video World Generation, and many more!
Read more → -
OpenAI begins a limited preview of GPT-5.6 Sol, a next-generation model with a layered safeguard stack and 700,000 GPU hours of automated red-teaming. Anthropic restores access to Claude Fable 5 and Mythos 5 after US export controls are lifted, and launches Claude Sonnet 5, its most agentic Sonnet model yet. AI summary
Read more → -
Advanced Claude Code Cost Tracking: How to Save 30% on Token Spend, Introducing Claude Tag, a paper on Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs, and many more!
Read more → -
GLM 5.2, Understanding Your Claude Code Spend: What’s Actually Driving the Cost, a paper on Data Journalist Agent: Transforming Data into Verifiable Multimodal Stories, and many more!
Read more →