TLDR AI 2026-06-17

Your bank has the data. But to actually use it — check cash position, reconcile transactions, understand burn — you have to export it, paste it somewhere else, and do the work outside your account.

Mercury Command closes that loop. It's AI built directly into Mercury that surfaces insights and takes action from your live account data. Ask what you need to know, then act on it instantly — follow up on an outstanding invoice, set a limit on a card, categorize a transaction — all in the same conversation. You approve every step. Command executes.

Try Mercury Command →

*Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column N.A., Members FDIC.

Z.ai launched GLM-5.2 with a 1 million-token context window, new reasoning controls, and support for long-horizon coding tasks across entire codebases. The company made the model available immediately to Coding Plan users and plans to release API access, chatbot support, technical details, and MIT-licensed open weights the following week. Z.ai positions GLM-5.2 as a coding-first upgrade focused on agentic software engineering, though it did not publish benchmark results at launch.

Android 17 introduced new AI-focused capabilities centered on AppFunctions and Android MCP, enabling apps to expose orchestratable tools that on-device agents can discover and execute. Google positioned the release as part of a broader shift toward an “intelligence system” with deeper AI integration across the platform.

With SiMa.ai's Palette Neat, developers build physical AI applications in days/hours - not months. Running your favorite AI model? Building custom, multimodal applications for robotics, automotive, drones, smart vision, or another physical AI project? Use AI to deploy physical AI! That's Neat. Learn more at the webinar.

Qwen-RobotWorld is a language-conditioned video world model that uses natural language as a unified action interface across robotics, navigation, driving, and other embodied domains.

Cursor Origin is an attempt at an agent-native GitHub competitor. The Git-compatible forge is built around the assumption that lots of AI agents will be cloning, branching, committing, rebasing, reviewing, and fixing failures in parallel. Cursor is trying to own the whole AI software factory. GitHub was built for human-scale development, but Origin was built for agent-scale.

OpenAI's Codex now supports Chrome DevTools Protocol, enabling live browser access to profile JavaScript performance and modify websites in real time. This early-stage feature, available via opt-in settings but excluded from the EEA, UK, and Switzerland, faces performance issues and requires careful prompting. Recent OpenAI moves, such as acquiring Ona, aim to create persistent cloud environments, highlighting a push toward an AI-driven web.

NVIDIA's Blackwell platform dominated MLPerf Training 6.0 benchmarks, achieving the fastest training times and the largest-scale training with 8,192 GPUs. Blackwell's NVLink and NVFP4 innovations enhance performance by efficiently routing MoE models and enabling rapid, large-scale training. Reliability features like the Reliability, Availability, and Serviceability Engine and NVIDIA Resiliency Extension ensure minimal downtime and faster recovery in large-scale AI environments.

Anthropic has paused the pricing changes it announced last month, just before they were set to take effect. The change would have treated Claude Agent SDK usage separately from standard Claude usage. Outside SDK usage will now be billed at Anthropic's prevailing API rates. Anthropic is working to update its plans to better support how users build with Claude subscriptions.

500,000 people have already signed up for TLDR Hardware, our new twice-weekly newsletter covering chips, robotics, energy, and devices. If you work in hardware and want to help curate it, send your LinkedIn or resume to hardware@tldr.tech (this email works now, so if your email bounced yesterday, please resend)!

Qualcomm is working on over 40 different AI wearable devices. It has announced a platform called Snapdragon Reality Elite for mixed-reality glass and the Scalable Turnkey AI-Ready toolkit, a combination of hardware models and a software stack for AI devices. The company is aiming to position itself as the foundational silicon layer for whatever is coming after the smartphone. It is experimenting with a range of potential form factors that can work with AI technologies.

Gartner engages with thousands of business leaders, IT executives, and technology providers, synthesizing a million proprietary data points. Get the latest insights in the Gartner AI hub

Winning teams at PMG's first-ever AI & Tech Sandbox hackathon get a portion of $500K in prize funding, a pathway to see their project deployed by global brands, and keep ownership of the IP. Register now

Crashes waste tokens when the provider connection lives inside the thing that crashed, so move it out.

GPT-Bidi-1 is a bidirectional audio model for ChatGPT's voice mode designed to listen and speak at once, absorb interruptions, and adjust mid-sentence.

The 3B parameter model put up coding benchmark scores in the same league as Claude Opus 4.5.

Microsoft's Phi Silica small language models are explicitly engineered to run locally on the Neural Processing Units of Windows Copilot+ PCs.