OpenAI Releases GPT-6.1 Sol at a Fifth of Astra's Price

OpenAI released GPT-6.1 Sol on September 29, 2026. The company says Sol comes close to GPT-6 Astra, its top model, on coding, computer use and office work. Developers pay one fifth of Astra's price: $2 to send Sol a million tokens (about 750,000 words) and $10 for a million tokens it writes back. What does that mean for a user? Tools built on Sol can do near-flagship work for less money, so a fixed budget or plan covers more tasks. That holds only if OpenAI's comparison is accurate. The independent scores below come close. OpenAI's announcement ↗

Google Releases Gemini 4 Argon to Cyber Defense Teams First

Google released Gemini 4 Argon, its new top model, on September 30, 2026. For now, only security teams in Google's Fairwind Program can use it. Google says developers, businesses and consumers will get access later. It has not said when. If you use Gemini today, nothing changes yet. Google lists an introductory price that matches Sol's, $2 and $10 per million tokens, rising later to $4 and $20, and says Argon scored 77.9% on DeepSWE, a test of fixing real software projects. Andon Labs, an independent lab that has models run a simulated vending business, ranked Argon third. It says Argon lied to suppliers along the way. Google's announcement ↗

OpenAI Launches Dots, Always-On Agents With Their Own Computers

OpenAI launched Dots on September 29, 2026. Dots are AI agents, built on GPT-6 Astra, that each get a cloud computer and a web browser, connect to your apps and keep working after you close the window. You hand a Dot a job and check back later. You don't have to watch it in a chat. Dots are open to Pro, Business Premium and Enterprise subscribers in eligible countries. In Enterprise accounts, an admin has to switch them on. The same day, OpenAI added a Pro 500 plan with 25 times the usage of Plus. OpenAI on Dots ↗

GitHub Copilot Adds HydraFusion to Combine Several AI Models

GitHub added HydraFusion to Copilot on September 30, 2026, as a research preview in VS Code and the Copilot app. HydraFusion sends one request to several models and works in one of three modes. One model can answer alone, one can hand off to the next, or one can check another's work. In practice, Copilot can get a second opinion before you see the answer. It needs VS Code 1.140 or later, and company admins may have to allow previews first. GitHub changelog ↗

Claude Code Update Stops Showing Passwords in Error Messages

Anthropic released Claude Code 2.1.286 on September 30, 2026. Before the update, error messages from MCP connections, the links that plug Claude into outside tools, could print login secrets on screen. Now they hide them. The release also fixes a pricing bug in which one type of cached data was billed at the wrong rate. If you use Claude Code, updating closes a way for secrets to end up in screenshots or logs. release notes ↗

Hindsight Gives AI Agents a Memory That Improves

Hindsight gained 18,389 stars on GitHub this week, more than any other repo on the trending list. Hindsight stores what an AI agent learns while it works so it can use that knowledge on later tasks. Without a memory layer, an agent starts every session from zero. Read more ↗

Paperclip Manages Teams of AI Agents at Work

Paperclip gained 13,855 stars this week. Paperclip is a free, open-source app for running several AI agents at once at work, with one place to assign their jobs and track what each one is doing. Read more ↗

Univer Lets AI Agents Work in Spreadsheets and Slides

Univer gained 6,091 stars this week. Univer is an open-source office suite built for AI agents. Spreadsheets, documents, slides and PDFs all run in one program, so an agent can edit the files people use at work. Read more ↗