What Is JARVIS-1 and Why Does It Matter for Open-World AI Agents? JARVIS-1 is an open-world multi-task agent that combines multimodal large language models with a growing memory system to complete over 200 different tasks in Minecraft using human-like control and observation spaces. Developed by Team CraftJarvis at Peking University and BIGAI, JARVIS-1 achieves a 5x reliability improvement over previous state-of-the-art agents on the challenging ObtainDiamondPickaxe task and near-perfect performance on short-horizon tasks like chopping trees. Its key innovation is a multimodal memory that blends pre-trained knowledge from the LLM with actual game survival experiences, enabling self-improvement without retraining. ...

The Kimi K3 Moment — What Happened on July 16, 2026 On July 16, 2026, Moonshot AI released Kimi K3, a 2.8-trillion-parameter open-weight model that instantly reshaped the competitive landscape of the large language model market. Built on a novel Mixture of Experts architecture with 16 out of 896 experts activated per token, K3 became the largest open-source model ever released and sent shockwaves through both Chinese and global AI markets. Within hours of the announcement, shares of competing Chinese AI companies plunged — Zhipu AI dropped 28% and MiniMax fell 16% — as investors recalibrated expectations for who leads the frontier. ...

Sx 2.0 is an open-source tool that turns a shared Dropbox folder into a complete AI skill distribution system for your team. Instead of emailing .claude.md files or maintaining a git repository of AI instructions, you drop a skill into a synced folder and every team member’s AI clients — Claude Code, Cursor, Copilot, Codex, Gemini, Cline, and Kiro — pick it up automatically. It is built by Sleuth (Dylan Etkin, former Atlassian) and is available under the Apache-2.0 license. ...

The Announcement — Moonshot AI Pauses New Subscriptions On July 18, 2026, Moonshot AI made an unusual announcement: it was temporarily suspending new subscriptions to its Kimi K3 model. The reason was not technical failure, regulatory pressure, or strategic retreat — it was overwhelming demand. In a post on X (formerly Twitter), the company stated that Kimi K3 “received far more love than expected” and that GPU capacity was “feeling it.” Within 48 hours of release, demand had pushed the service close to its capacity limits, forcing Moonshot to take the rare step of pausing new sign-ups to protect the experience of existing subscribers. ...

Grok Build is xAI’s open-source, terminal-first AI coding agent that uses Grok 4.5 to plan, edit, test, review, and ship software directly from the command line. Released in mid-2026, it competes directly with Claude Code, Codex CLI, and Cursor by offering a skill-based agent framework with MCP server support, AGENTS.md configuration, and a terminal user interface — but it has also drawn significant scrutiny over an undisclosed full-repository upload mechanism that transmits entire codebases to Google Cloud Storage without explicit user consent. ...

Introduction — What Is Sim Studio? Sim Studio (now rebranded simply as “Sim”) is an Apache-2.0 licensed, open-source visual agent workflow builder that lets you design, simulate, and deploy multi-agent AI systems through a drag-and-drop GUI. Launched in January 2025 and backed by Y Combinator’s X25 batch, Sim has grown to over 29,000 GitHub stars in 18 months, positioning itself as the leading fully open-source alternative to n8n for AI-native workflow automation. Unlike fair-code competitors, Sim offers unlimited free self-hosting, a natural-language control plane called Mothership, and 1,000+ native integrations across AI models, communication tools, and databases. ...

Finterm.ai is a CLI-first financial data platform designed specifically for AI coding agents like Claude Code, ChatGPT, and Cursor. Instead of a traditional GUI dashboard, it delivers stock prices, SEC filing diffs, options sentiment, insider trades, and deep ticker research directly into your agent’s command line — making it the closest thing to a Bloomberg Terminal for AI-powered development workflows. What Is Finterm.ai? Finterm.ai is a financial data API wrapped in a command-line interface, purpose-built for AI agents rather than human traders. Founded by Kam and Josh under DXDT Labs Incorporated, the product launched as a Show HN on Hacker News on July 13, 2026. The founding story is rooted in a real trading experience: Kam made a 16% return on a Popmart Labubu short thesis using LLM-assisted research, but the process was painful — manually fetching SEC data, copy-pasting into GPT, and juggling multiple chat windows. Finterm was built to eliminate that friction. ...

Introduction — The Year AI Agent Benchmarks Broke If you have been following AI agent benchmarks in 2026, you have likely seen headline numbers that look too good to be true. A tool called BenchJack scored 100% on SWE-bench Verified, SWE-bench Pro, and Terminal-Bench without solving a single task — most runs never even called a language model. OpenAI retired SWE-bench Verified after discovering that 59.4% of hard failed tasks had broken test cases rejecting functionally correct patches. And the same frontier model scored 64.7% on one wrapper and 57.5% on another — a 7.2-point gap from scaffolding alone. This is the state of exploiting AI agent benchmarks in 2026: a system where the incentives to publish high scores have outpaced the rigor of the evaluations themselves. ...

The 2026 LLM landscape is defined by benchmark fragmentation and a widening gap between frontier performance and cost. No single model dominates across all tasks: GPT-5.6 Sol leads reasoning with 94.6% on GPQA Diamond, Claude Opus 4.8 excels at long-context coding with a 1M token window, DeepSeek V3.2 delivers ~80% of frontier performance at under 10% of the cost, and Gemini 3.5 Flash is the fastest measured model at 284.2 tokens/second. The key to choosing the right model and API provider is understanding which benchmarks measure your actual use case. ...

Introduction — The Benchmark Illusion For the last two years, the AI industry has been racing to top leaderboards on agent benchmarks like SWE-bench, WebArena, and GAIA. These scores drive funding rounds, product launches, and enterprise procurement decisions. But a landmark study from Berkeley RDI reveals a devastating truth: every single major AI agent benchmark can be exploited to produce near-perfect scores without the agent solving a single real task. The study, led by Hao Wang, Qiuyang Mang, Alvin Cheung, Koushik Sen, and Dawn Song, introduces BenchJack — an automated vulnerability scanner that achieved 100% scores on eight benchmarks using nothing more than environment manipulation, configuration leakage, and broken validation logic. The paper, titled “Do Androids Dream of Breaking the Game?” (arXiv 2605.12673), demonstrates that the current state of AI agent evaluation is fundamentally broken, and the problem is far more urgent than most in the industry realize. ...