A comprehensive review and comparison of ten open-weight large language model releases from January to February 2026, including Arcee Trinity, Kimi K2.5, Step 3.5 Flash, Qwen3-Coder-Next, GLM-5, MiniMax M2.5, Nanbeige 4.1, Qwen3.5, Ling 2.5, Tiny Aya, and an update on Sarvam. The article focuses on architectural similarities and differences, highlighting trends like hybrid attention, multi-token prediction, and mixture-of-experts.

Researchers propose a framework to measure AI agent reliability, decomposing it into 12 dimensions across four categories. Testing 14 models over 18 months reveals rapid capability improvements but only modest reliability gains, calling for reliability-specific optimization.

Seedream 5.0 brings multi-step reasoning, example-based editing, and deep domain knowledge to image generation. Here's what you should know.

This issue covers the role of measurement in AI governance, LLMs in nuclear crisis simulations, China's ForesightSafety Bench, and the LABBench2 benchmark for AI in science.

Setup OpenClaw in under two minutes with a single Ollama command.

In March 2025, we dropped a statement-bomb in the “Relevance Feedback in Information Retrieval” article and then went silent. We claimed that even though the information retrieval research field has proposed many useful mechanisms for increasing the relevance of search results, none of them made it to the neural search industry, simply because these approaches are not scalable. Certainly, there are methods widely used to improve the relevance of retrieved results: query rewriting, for example. Yet none of the vector search solutions out there have tried to use the possibilities that come with full access to the vector search index: traversing it in the direction of relevance, instead of guessing where to shoot the query to the vector space.

This essay argues that rational people don’t have goals, and that rational AIs shouldn’t have goals. Human actions are rational not because we direct them at some final ‘goals,’ but because we align actions to practices. It proposes ‘eudaimonic rationality’ as a framework for AI alignment that shares a ‘type signature’ with human practice-based logic, addressing safety properties like transparency, corrigibility, and more.

AI usage has shifted from chatbots to agents. This guide explains the three factors to consider—models, apps, and harnesses—and provides a detailed overview of the current landscape, including GPT, Claude, Gemini, and specialized tools.

Ollama now supports subagents and web search in Claude Code. No MCP servers or API keys required. Subagents can run tasks in parallel, keeping context clean. Web search is built-in via the Anthropic compatibility layer.

Applying the AI as Normal Technology framework to legal services, this article argues that advanced AI will not by default help consumers achieve desired legal outcomes at lower costs due to three bottlenecks: regulatory barriers, adversarial dynamics, and human involvement. It also discusses potential institutional reforms.

Replicate's MCP server can now be discovered via the official MCP Registry.

OpenClaw is a personal AI assistant that connects your messaging apps to local AI coding agents, all running on your own device for privacy.

This famous aphorism is neither true nor useful

In an experiment at the University of Pennsylvania, MBA students used AI tools to create startups from scratch in four days, demonstrating how AI accelerates entrepreneurship. The article explores an equation for agentic work and how effective delegation (management skills) improves AI success rates.

Inference-time scaling is one of the most effective ways to improve answer quality in deployed LLMs. This article categorizes various inference-time scaling techniques and provides an overview of recent papers, including chain-of-thought prompting, self-consistency, best-of-N ranking, rejection sampling with a verifier, self-refinement, and search over solution paths. The author shares personal experiments from drafting a book chapter on the topic.

Ollama introduces `ollama launch`, a new command that sets up and runs coding tools like Claude Code, OpenCode, and Codex with local or cloud models, without needing environment variables or config files.

San Francisco-based cloud platform Railway, which has amassed two million developers without marketing spend, announced a $100 million Series B round. With sub-second deployments, vertically integrated data centers, and per-second billing, it positions itself as a critical infrastructure startup in the AI era.

Anthropic's Claude Code pricing sparks developer backlash, while Block's open-source AI agent Goose offers similar functionality for free, running locally with privacy and offline support. This article analyzes Claude Code's rate limit controversy, Goose's features and setup, and the trade-offs in model quality, context window, speed, and tooling maturity.

Listen Labs, an AI-powered customer interview platform, raised $69M in Series B funding at a $500M valuation. The company gained attention with a creative billboard hiring challenge and has since grown annualized revenue 15x. Its platform conducts in-depth interviews via AI, replacing traditional surveys and focus groups, and is used by Microsoft, Sweetgreen, and others. The company plans to expand into synthetic customers and automated decision-making.

Ollama v0.14.0+ now supports the Anthropic Messages API, enabling tools like Claude Code to work with open-source models. Run locally or connect to cloud models via ollama.com.

Open models can be used with OpenAI's Codex CLI through Ollama. Codex can read, modify, and execute code in your working directory using models such as gpt-oss:20b, gpt-oss:120b, or other open-weight alternatives.

Filter the list predictions API endpoint to show only predictions created through the web interface using the source=web parameter. Note that results are limited to the last 14 days.

Salesforce has launched a fully rebuilt Slackbot powered by Anthropic's Claude, transforming it from a simple notification tool into an AI agent that searches enterprise data, drafts documents, and takes actions. The new Slackbot is available to Business+ and Enterprise+ customers at no extra cost. Internally, 80,000 Salesforce employees tested it, reporting high satisfaction and time savings. Pilot customers like Beast Industries report saving up to 90 minutes per day. Slackbot competes with Microsoft Copilot and Google Gemini, with Salesforce positioning it as a 'super agent' for the enterprise. The rollout begins today, with mobile availability by March.

AI drug discovery startup Converge Bio has raised $25 million in a Series A round led by Bessemer Venture Partners, with participation from TLV Partners, Saras Capital, Vintage Investment Partners, and executives from Meta, OpenAI, and Wiz. The company uses generative AI trained on molecular data to accelerate drug development for pharma and biotech firms.

Anthropic releases Cowork, a new AI agent that extends Claude Code's capabilities to non-technical users via a folder-based desktop interface. Built in about a week and a half, largely using Claude Code itself, Cowork is available as a research preview to Claude Max subscribers on macOS. It can read, edit, and create files, integrate with connectors and browser automation, and includes safety warnings about potential file deletion and prompt injection risks.

Researchers have developed a framework to evaluate and optimize imaging systems based on mutual information, predicting performance across four domains and enabling efficient design without task-specific decoders.

New AI coding tools like Claude Code demonstrate remarkable autonomous capabilities, completing complex tasks and self-correcting errors. The article explores features like long-running autonomy, context compaction, skills, and sub-agents, and discusses their profound impact on programming. While currently focused on coders, these tools hint at broader AI applications in knowledge work.

Nous Research, the open-source artificial intelligence startup backed by crypto venture firm Paradigm, released a new competitive programming model on Monday that it says matches or exceeds several larger proprietary systems — trained in just four days using 48 of Nvidia's latest B200 graphics processors.

Boris Cherny, creator of Claude Code at Anthropic, shared his personal terminal workflow on X, sparking a viral discussion. His approach includes running 5 Claude agents in parallel, using the Opus 4.5 model, maintaining a CLAUDE.md file for error lessons, and leveraging slash commands with subagents for automation. This workflow turns coding into a real-time strategy game, enabling a single developer to achieve the output of a small engineering team.

Spellar AI releases version 3.0 with cross-meeting memory, enabling users to recall discussions and decisions from past calls. The update also introduces client/project organization, templates, and choice of AI models.