Apple's Xcode now has direct integration with the Claude Agent SDK, giving developers the full functionality of Claude Code for building on Apple platforms, from iPhone to Mac to Apple Vision Pro.

Google launched a first-of-its-kind nationwide randomized study with Included Health to evaluate AI in a real-world virtual care setting and better understand its capabilities and limitations

Google launched Institutional Review Board (IRB) approval, a prospective consented nationwide randomized study to assess AI in a real-world virtual care setting. This new research will build upon our foundational research on the use of AI for diagnostic and management reasoning, personalized health insights and navigating health information.

This study is informed by years of foundational research across Google, investigating the capabilities required for a helpful and safe medical AI.

Meet Q Labs a research lab focused on solving generalization.

Alongside others (SSI, Flapping Airplanes), Q Labs see data efficiency as the key problem, but they're taking an unconventional approach to solve it: a new learning algorithm approximating Solomonoff induction.

Why Solomonoff Induction? It's provably optimal for prediction. The idea is simple: search for all programs that fit the data and favor low-complexity ones.

Since it's uncomputable, they're building a practical approximation in the context of neural nets.

Google put out an article on accelerating research with Gemini.

Mistral Introduced Voxtral Transcribe 2, next-gen speech-to-text models

SOTA transcription, speaker diarization, sub-200ms real-time latency.

Voxtral Realtime is built for voice agents and live applications. Its natively streaming architecture delivers latency configurable to sub-200ms. And at 480ms, it stays within 1-2% WER of our offline model.

Mistral released the model as open weights under Apache 2.0.

The demo is worth a try - ignore the "No microphone found" message, clicking "Record" and allowing your browser to use a microphone fixes that. It transcribes very accurately in almost real-time. It's really impressive.

OpenAI introduced Frontier a new platform that helps enterprises build, deploy, and manage AI coworkers that can do real work.

Frontier gives agents the same skills people need to succeed at work:

- Understand how work gets done
- Use a computer and tools
- Improve quality over time
- Stay governed & observable

Also pair OpenAI Forward Deployed Engineers with your team, working side by side to develop your best practices to build and run agents in production.

Anthropic Introduced Claude Opus 4.6, It’s first Opus-class model with 1M token context in beta

Opus 4.6 plans more carefully, sustains agentic tasks for longer, operates reliably in massive codebases, and catches its own mistakes.

Opus 4.6 is SOTA on several evaluations including agentic coding, multi-discipline reasoning, knowledge work, and agentic search.

Anthropic also shipping new features across Claude in Excel, Claude in PowerPoint, Claude Code, and API to let Opus 4.6 do even more.

Claude in Excel now handles long-running and harder tasks with improved performance.

It can plan before acting, support richer functionalities like conditional formatting and data validation, and handle multi-step changes in one pass.

On Claude Code Anthropic introduced agent teams.

Spin up multiple agents that coordinate autonomously and work in parallel—best for tasks that can be split up and tackled independently.

Agent teams are in research preview

General framework of AI agents.

What if a single framework could unify every AI agent, from software chatbots to physical robots?

Dr. Hang Li from ByteDance proposes exactly that in a new JCST paper.

It's a universal blueprint where agents use LLMs as their "brain" for reasoning, are built via reinforcement learning, and operate using tools & long-term memory to complete tasks.

This general framework outperforms fragmented approaches, providing a unified theory for agent development across both software and hardware domains.

$11T by 2030. This is the prediction ARK Investment estimates on-chain assets could grow to ~$11T by 2030, driven by deposits, public equities, credit, and funds moving on-chain.

This is where we are today:

• Stablecoin supply crossed $300B in 2025, with real transaction volumes now competing with legacy payment rails

• Tokenised real-world assets reached ~$19B, led by Treasuries and commodities

• Ethereum hosts the majority of that on-chain value

This is what will fuel the growth:
• Deposits moving on-chain for faster settlement and global liquidity
• Public equities and funds reducing issuance and operational costs via tokenisation
• Credit markets adopting programmable settlement and collateral workflows
• Banks, asset managers, fintechs, and payment networks actively launching on-chain rails
• Public blockchains increasingly used as back-end infrastructure, not front-end products

ARK Investment Management LLC wording is telling:

“Ethereum remains the preferred blockchain for on-chain assets.”

$19B today.
$11T by 2030.

KPMG told its auditor, Grant Thornton UK, it should pass on cost savings from the rollout of AI and threatened to find a new accountant if it did not agree to a significant fee reduction, the people said.

Waymo introduced World Model a frontier generative mode for large-scale, hyper-realistic autonomous driving simulation built on Google DeepMind’s Genie 3.

Meta is preparing to get new avocado models, Manus browser agent and integration with Openclaw

What’s new?

- Meta AI website got migrated to the new stack while retaining the same user experience.

- New effort selector, email, and calendar connectors are already available to users.

- Memory and Projects are in the works.

- Avacado and Avacado Thinking models have been spotted in testing. The router still redirects tothe Llama model.

- Meta is testing different models from other providers underneath, including Gemini 3 Pro preview, Claude Sonnet 4.5 and GPT-5.2. This part was likely inherited from Manus AI but used only internally.

- Scheduled tasks feature is under development.

- A new model named Sierra is being tested to power the upcoming browser agent. This likely will be the same agent used in Manus AI currently.

- Big Brain mode is in the works, where multiple model responses will be combined into a final answer.

- OpenClaw integration has been spotted to let Meta AI connect to your OpenClaw gateway.

In short, Meta is on the path to slowly transform into Manus. OpenClaw integration might be big, but something that other Labs may easily adapt as well. We are yet to know if Avacado's performance will surpass recently released models, where Avacado is expected to be released around this spring.

Chinese new year is in less than 10 days. This is DeepSeek or Zhipu GLM-5?

OpenRouter announced a new “stealth” large language model, Pony Alpha, described as a next-generation foundation model excelling in coding, reasoning, and roleplay tasks, and optimized for agentic workflows with precise tool-calling.

It is free. The provider logs all prompts and completions for this model, which may be used to improve the model.

The model is free on OpenRouter, though all prompts and completions are logged to improve it.

Multiple Tech PhDs and Silicon Valley entrepreneurs speculate it could be DeepSeek-V4, Zhipu GLM’s new model, or Grok 4.2/Claude 5, with the “Pony” name and Year of the Horse hinting at a Chinese origin.

OpenRouter partner Kilo Code suggested in a blog that Pony Alpha is “a special evolution of a popular open-source global model,” making DeepSeek-V4 or Zhipu GLM-5 the most likely candidates.

Zhipu up over 40% in Hongkong at one point, hitting new peak on AI optimism.

Meet EchoJEPA is the first world model for medical video

• 18M echocardiograms
• 300K patients
• Learns heart dynamics — not imaging noise

EchoJEPA discards what’s unpredictable and locks onto what matters clinically:
- chamber geometry
- wall motion
- valve dynamics

The results (frozen encoder, no fine-tuning):
• 20% ↓ error in LVEF
• 17% ↓ error in RVSP
• 79% accuracy with 1% labels (vs 42% for baselines w/ 100%)
• 2% degradation under acoustic artifacts (vs 17%)
• Zero-shot pediatric transfer beats all fine-tuned models

GitHub.

Google DeepMind introduced a new paper on learning temporally abstract world models and policies (options).

Key idea:

1.use LLM to propose features for a factorized product of experts world model;

2. use this to predict abstract world state after each macro action to help RL explore.

Google, UC Berkeley and an international team of researchers present Aletheia, a math research agent built on Gemini

The system uses AI to systematically scan hundreds of complex conjectures, filtering through potential proofs with natural language verification before sending the best candidates to human experts for final review.

The team resolved 13 "open" problems from the Erdős database, generating 4 brand-new solutions and identifying 9 others that were actually solved in obscure corners of existing literature.

Bytedance dropped advanced video generation model

Seedance 2.0 has:
— native audio gen (lipsynced speech + music)
— drastic step up from Veo 3.1 / Sora 2 in quality
— supports multimodal input
— 2k resolution

Goes beyond cinematic video, and can do product demos as well. And it's really hard to tell it's AI.

The PaddleOCR Document Parsing Skill is now live on ClawHub, ready to plug directly into OpenClaw workflows.

Instead of deploying OCR services or wiring APIs, developers can now invoke PaddleOCR as a standardized composable Skill node — embedding document understanding directly into Agents and automation pipelines.

Built on PaddleOCR-VL-1.5, the Skill delivers
1. Multi-format parsing (PDF, JPG, PNG, BMP, TIFF)
2. Layout analysis — text, tables, formulas, headers
3. 110+ language coverage
4. Structured Markdown output preserving hierarchy

No deployment. No wrappers. Just configuration — and build your document intelligence chain inside OpenClaw.

What if your model could learn from its own drafts during RL training?

NVIDIA introduced iGRPO: Iterative Group Relative Policy Optimization.

Researchers add a self-feedback loop to GRPO: the model drafts multiple solutions, picks its best one, then learns to refine beyond it.

Core idea:
Stage 1 → explore and select your strongest attempt. Stage 2 → condition on that attempt and beat it.

Same scalar reward. No critics, no generated critiques, no verification text. The best draft is the only feedback the model needs.

Results across 7B / 8B / 14B models:

• Nemotron-H-8B-Base-8K: 41.1% → 45.0% (+3.96 over GRPO)

• DeepSeek-R1-Distill-Qwen-7B: 68.3% → 69.9%

• OpenMath-Nemotron-14B: 76.7% → 78.0%

• OpenReasoning-Nemotron-7B on AceReason-Math: 85.62% AIME24 / 79.64% AIME25

The same two-stage wrapper also improves DAPO and GSPO. It's not tied to GRPO at all.

Google introduced DialogLab a new open-source prototyping framework, uses a human-in-the-loop control strategy to achieve realistic human-AI group simulation, offering a necessary alternative to fully autonomous agents.

Evaluations with domain experts found that its "Human Control" mode (where you can edit, accept, or dismiss real-time AI suggestions) was preferred in realism, effectiveness, and engagement.

DialogLab transforms dialogue design from rigid scripts to spontaneous, adaptable group dynamics.

This new research introduces Agyn, an open-source multi-agent platform that models software engineering as a team-based organizational process rather than a monolithic task.

The system configures a team of four specialized agents: a manager, researcher, engineer, and reviewer. Each operates within its own isolated sandbox with role-specific tools, prompts, and language model configurations. The manager agent coordinates dynamically based on intermediate outcomes rather than following a fixed pipeline.

What makes the design interesting?

Different agents use different models depending on their role. The manager and researcher run on GPT-5 for stronger reasoning and broader context. The engineer and reviewer use GPT-5-Codex, a smaller code-specialized model optimized for iterative implementation and debugging. This mirrors how real teams allocate resources based on task requirements.

The workflow follows a GitHub-native process. Agents analyze issues, create pull requests, conduct inline code reviews, and iterate through revision cycles until the reviewer explicitly approves. No human intervention at any point. The number of steps isn't predetermined. It emerges from task complexity.