Eigent — the first open source multi-agent workforce on your desktop.

Eigent is a team of AI agents collaborating to complete complex tasks in parallel.

It brings together specialized agents, developer, search, document, multi-modal, each designed to work in parallel and adapt to your needs.

Eigent is built on CamelAI open-source multi-agent infrastructures.

It supports:
- Running parallel tasks
- Custom workers
- Cloud version or "Bring Your Own Key" (BYOK)
- Local model deployment
- Human-in-the-loop feedback
- Model Context Protocol (MCP) tools
- Secure self-hosting
- Enterprise-level security

Eigent supports multiple deployment options:

- Cloud version with instant access and managed infrastructure
- Community edition for local hosting and customization
- Enterprise edition with SLAs, auditability, and scale

Coinbase and JPMorgan have partnered to crypto access for over 80 million Chase customers, introducing three 3 methods:

- converting Chase Ultimate Rewards to USDC,
- funding Coinbase accounts with Chase credit cards,
- direct bank integration.

The integration of Ultimate Rewards to USDC offers a novel entry point, while credit card funding and direct bank links streamline transactions, potentially boosting adoption rates among mainstream users.

BlockDL a free & open-source GUI that lets you visually design Keras neural networks and learn ML.

Langchain introduced Deep Agents

Team created a new Python package which makes it easy to build your own Deep Agents.

The core algorithm for Deep Agents is actually the same - it’s an LLM running in a loop calling tools. The difference is:

1. Planning tool
2. Sub agents
3. File system
4. A detailed system prompt (prompting is not dead!)

Google Introduced AlphaEarth Foundations an AI model that integrates petabytes of satellite data into a single digital representation of Earth.

It'll give scientists a nearly real-time view of the planet to incredible spatial precision, and help with critical issues like food security, deforestation & water resources

Deep cogito released 4 hybrid reasoning models of sizes 70B, 109B MoE, 405B, 671B MoE under open license.

The models are built on Deep cogito’s work on building superintelligence using Iterated Distillation and Amplification (IDA). In particular, team scale the model’s intelligence prior by the model internalizing the reasoning process using iterative policy improvement, rather than simply searching longer at inference time.

This seems to be a novel scaling paradigm where the models develop more “intuition”, and serves as a strong proof of concept for self-improvement. Since the Cogito models develop a better intuition of the trajectory to take while searching at inference time, they have 60% shorter reasoning chains than Deepseek R1.

Meta introduced MetaCLIP 2

The effort addresses long-standing challenges:

1. large-scale non-English data curation pipelines are largely undeveloped,

2. the curse of multilinguality, where English performance often degrades in multilingual CLIP compared to English-only CLIP.

With a complete recipe for worldwide CLIP—spanning data curation, modeling, and training—we show that English and non-English worlds can mutually benefit and elevate each other, achieving SoTA multilingual performance.

GitHub.

The International Monetary Fund’s new working paper breaks empirical ground by mapping $2 trillion in global #stablecoin transactions during 2024 across five regions (US, Europe, APAC, MENA, LATAM) using a novel AI- and machine learning–based methodology.

By analyzing over 138 million on-chain transactions and nearly 6 million wallet domain names, the study reveals that while stablecoin volumes are highest in North America ($633bn) and Asia-Pacific ($519bn), their macroeconomic significance is greatest in Latin America (7.7% of GDP) and Africa (6.7%).

Crucially, the US emerges as the dominant net exporter of stablecoins ($54bn net outflows), with flows intensifying during periods of dollar strength, suggesting that stablecoins now serve as an agile instrument for meeting global dollar demand, akin to #Eurodollars but operating at #blockchain speed.

The March 2023 US banking crisis, triggered by the collapse of several regional banks servicing #crypto firms significantly disrupted stablecoin flows originating from North America, as evidenced by a sharp decline in on-chain transaction volumes during the crisis period.

Methodologically, the paper also challenges existing datasets, showing that their reliance on web traffic and VPN-free assumptions underestimates stablecoin use in regions like China by a factor of 5.5.

Instead, the IMF’s region-classification model (trained on 350,000 wallets) captures behavioral and time-zone-specific transaction patterns, offering a more robust lens into crypto capital flows.

Another open model from a Chinese AI lab outperforms closed ones

XBai o4 beats OpenAI o3-mini and confidently beats Anthropic's Claude Opus.

It is using parallel thinking, something similar to Gemini DeepThink, o1 pro, o3 pro.

Apache 2.0 license and available on HF.

Meta FAIR Chemistry team announced FastCSP, a workflow that generates stable crystal structures for organic molecules.

This accelerates material discovery efforts and cuts down the time to design molecular crystals from months to days.

The workflow will be available soon here.

AI can now design perfect custom chips for other AIs 9.5x faster.

Researchers presenting at the International Conference on Computer-Aided Design developed a framework that automates ASIC chip optimization for LLMs. And they open sourced it on Github.

The system, called Coflex, gives designers a menu of optimal choices based on maximizing accuracy, speed or power consumption.

It navigates a search space with over 10¹⁸ hardware and software configurations.

Instead of slow, exhaustive testing, it uses Sparse Gaussian Processes (SGP). SGP creates an intelligent, probabilistic "map" of the entire space using only a small set of representative "landmarks" called inducing points.

This allows Coflex to accurately predict the performance of untested designs. It then simultaneously optimizes for conflicting goals like minimizing error rate and power consumption (Energy-Delay-Product) to identify the Pareto front of ideal trade-offs.

The key innovation is using SGP to reduce the computational complexity of multi-objective Bayesian optimization from O(n³) to near-linear O(nm²), solving the scalability bottleneck in Hardware-Aware Neural Architecture Search (HW-NAS).

GitHub.

Google DeepMind Introduced Genie 3, a SORA world model that generates interactive worlds from text, enabling real-time interaction at 24 fps with minutes-long consistency at 720p

One emergent capability is long-term consistency, especially because we don’t use any explicit 3D representations or priors.

Simply training the model to generate the next frame auto-regressively teaches it to maintain physical consistency across time.

thThefuture iterations of models like Genie 3 will have a significant impact on accelerating robotics and real-world AI.

An agent pursuing a goal in an environment generated this model.

OpenAI released gpt-oss: SOTA open-weight language models that deliver strong real-world performance. Runs locally on a laptop

Also Anthropic launched sota coding with Claude Opus 4.1

Claude Opus 4.1, an upgrade to Claude Opus 4 on agentic tasks, real-world coding, and reasoning.

Meet RoboMonkey is a framework for synthetic data generation + scaling test time compute for VLAs

Turns out generation (via repeated sampling) and verification (via training a verifier on synthetic data) works well for robotics too.

GitHub.
Datasets and models.
Serving engine.

Salesforce introduced CoAct-1 — a hybrid agent that elevates coding to a first-class action alongside GUI manipulation.

On OSWorld, CoAct-1 achieves a new SOTA score of 60.76%, becoming the first CUA agent to cross the 60-point mark.

Takeaways:
- Treat code as an action, not just a tool call.
- Hybrid action space (code + GUI) reduces error accumulation and boosts reliability.
- New SOTA on OSWorld with better efficiency and broader applicability.

Paper.

Google Introduced DeepPolisher is a new open-source method to improve genome assembly accuracy. It reduces indel errors by 70% and total assembly errors by 50%.

Microsoft presented Agent Lightning

Enables seamless agent optimization for any existing agent framework (e.g. LangChain) with any optim framework (e.g. DSPy) without any modifications to the agent code.

Paper.
Repo.