Ecosystem

One API call and the agent runs: task, model, tools, environment. No markup on top of tokens. Anthropic staked out the same ground with Managed Agents, on the same day.

OpenAI reworked image generation: up to 50 percent lower latency, reference photos that stay recognizable, and Flare and Sunburst for developers.

In a blog post, OpenAI's finance chief lays out the numbers she'll likely be taking to investors: a billion weekly users, 2.5 million businesses — and a research team burning 3.1 agent-workdays for every human workday.

With version 1.30.0 of the ant CLI, agents, skills, environments, memory stores, and deployments can be described as files in your repo and synced with the API in one command. A lockfile remembers what already exists. If you know Terraform, you'll feel right at home.

Astra uses a technique called 'recurrent depth': the model runs a query through itself several times instead of writing out its reasoning step by step. That makes the chain of thought harder to read. Redwood Research and Zvi Mowshowitz warn of a race to the bottom.

The GPT-5.6 models are now available in Kiro, AWS's coding agent. OpenAI says Terra completes tasks there roughly 82 percent cheaper — thanks to Kiro's spec-driven approach.

OpenAI published the first measurements for Jalapeño, its first in-house inference chip. Up to 1.9x more compute per watt and much lower latency — and AI helped design the thing itself.

Of all people, Linus Torvalds fixed a nasty graphics bug in the Linux kernel with help from an AI. In the end it was a single line. But the interesting part is what he wrote about the AI.

Payments company Stripe is acquiring OpenRouter – the gateway that makes hundreds of AI models reachable through a single endpoint. A fat deal that says a lot about where the LLM market is heading.

OpenAI turns the price dial on its top model again: $4 instead of $5 per million input tokens, $20 instead of $30 on output. For three months, and pretty clearly meant as a message to Anthropic's Opus 5.

OpenAI is bringing ads to ChatGPT in Europe. Starting next week in 31 countries, Germany included. Only Free and Go users are affected - if you pay, you stay ad-free.

OpenAI shipped an Apple Messages plugin for the ChatGPT desktop app. On Apple Silicon Macs, ChatGPT can search iMessage, SMS, and RCS, catch you up on threads, and prepare replies – sending only after you approve.

OpenAI handed ChatGPT a few solid little upgrades: interactive quizzes right inside the chat, switchable memory per project, and the Think button now on Free too. Nothing earth-shaking – but you'll feel it day to day.

OpenAI gives an early look at a new service tier that runs GPT-5.6 Sol up to 14 times faster. Not a new model, just raw speed – powered by Cerebras chips.

The Linux community has been begging for a while, and now the dedicated ChatGPT app is here, Codex included. Anthropic, though, shipped the Claude desktop for Linux a month earlier.

AMD is acquiring Taalas, a startup that casts trained models directly into silicon. A demo chip hit over 16,000 tokens per second. Here is why that matters for the inference cost of Claude and friends.

Free and Go users get unlimited text chats and GPT-5.6 Luna as the new default model. Add a Think button for hard questions - and Sol gets sharpened for paying users.

Meta unveiled Muse Code — an AI agent for large codebases that handles complete engineering tasks. It runs on Muse Spark 1.2 and fans big jobs out into parallel sub-agents. The pitch: cheaper than Codex and Claude Code.

New reports sharpen the picture of OpenAI's first hardware product: a portable, donut-shaped AI speaker about the size of a hockey puck, designed with Jony Ive. With cameras, moving parts, and a price around 300 to 400 dollars — planned for 2027.

After 27 years, chief scientist Jeff Dean is leaving Google to launch a startup called Discovery Loop with Sanjay Ghemawat. Demis Hassabis becomes Alphabet's Chief Scientist, and CTO Koray Kavukcuoglu takes over DeepMind. The stock fell more than 4 percent.