AI NewsSeptember 23, 2026

Meta's Muse AI agent hit #1 on the US App Store and struck deals with PayPal, Expedia, Shopify, and Instacart to handle checkout and travel booking directly, though Amazon has blocked it from shopping on its site.
Meta's Muse AI agent reached #1 on Apple's US App Store on September 22, just two weeks after its September 8 launch. New partnerships let it check out at PayPal merchants worldwide, book hotels through Expedia, and build grocery carts through Instacart.
Meta also opened its connector platform to more third-party developers the same day, pushing Muse from a conversational assistant toward a transaction layer. Not everyone's on board: Amazon has blocked Muse from shopping on its site, and Charles Schwab shares dropped over 6%.
Key Capabilities:

• Fast climb: Hit #1 on the US App Store just two weeks after its September 8 launch.
• Transaction-ready: PayPal, Shopify, Expedia, and Instacart integrations turn requests into checkouts.
• Not universal: Amazon has blocked Muse from shopping on its site.

Developer Atai Barkai released OpenMuse, a self-hostable personal AI assistant with browser, terminal, and file access that works with any agent harness, built on CopilotKit and AG-UI.
Developer Atai Barkai released OpenMuse on September 22, an open-source, self-hostable personal assistant built with CopilotKit and AG-UI. It includes computer use across a browser, terminal, and files, connectors for personal apps, and idea and goal tracking, available for both mobile and web.
The project is MIT licensed and meant to be cloned and customized rather than used as-is; it's currently alpha, with some features still requiring extra setup. It arrives the same week Meta's own Muse app surged to the top of the App Store.
Key Capabilities:

• Self-hosted: Runs on your own infrastructure, no dependence on Meta's servers.
• Computer use: Persistent browser and terminal access, plus file handling between agent and app.
• Bring your own harness: Works with any agent backend, not locked to one AI provider.

Moonshot AI rebranded Kimi WebBridge into Kimi Browser Extension, adding a chat sidebar and the ability to record repetitive web tasks once and save them as reusable skills.
Moonshot AI renamed Kimi WebBridge to Kimi Browser Extension on September 22, adding a browser sidebar where you can chat with Kimi to navigate sites, fill out forms, and get things done. It runs locally, so login sessions and page content never leave your device.
For repetitive tasks, you can record your steps once and save them as a skill Kimi reuses next time. It's available now on the Kimi website and the Chrome Web Store, working with agents like Kimi Code, Claude Code, Cursor, and Codex.
Key Capabilities:

• Skill recording: Record a repetitive web task once, save it, and Kimi replays it later.
• Runs locally: Login sessions and page content stay on your device, never sent to the cloud.
• Broad agent support: Works with Kimi Code, Claude Code, Cursor, and Codex.

OpenAI launched GPT-6 Sol and Luna, cheaper models built on GPT-6 Astra's improvements, at half the API price of their GPT-5.6 predecessors, with Sol making about half as many factual mistakes.
OpenAI launched GPT-6 Sol and Luna on September 22, extending GPT-6 Astra's gains to cheaper tiers. Sol costs $2/$10 per million tokens, half of GPT-5.6 Sol's price, while Luna comes in at $0.10/$0.50. Sol makes about half as many mistakes as its predecessor.
Luna is optimized for fast, high-volume tasks, while Sol carries more reasoning capability for coding and computer use. Sol now matches or beats Claude Fable 5.1 on some benchmarks, though Anthropic released a competing Opus 5.5 the very same day.
Key Capabilities:

• Half the price: Sol and Luna cost 50% less than their GPT-5.6 counterparts.
• Fewer mistakes: Sol makes about half as many factual errors as GPT-5.6 Sol.
• Split by use case: Sol handles complex reasoning; Luna targets fast, high-volume tasks.

Anthropic launched Claude Opus 5.5, performing at the level of Claude Fable 5.1 on most tasks while costing 40% less to run than Opus 5, with removed five-hour usage caps for subscribers.
Anthropic launched Claude Opus 5.5 on September 22, the first model in a new 5.5 family. It performs at the level of Claude Fable 5.1 for most tasks, generates responses over 30% faster than Opus 5, and costs 40% less on typical workloads.
Anthropic is also removing five-hour usage caps for Pro, Max, Team, and Enterprise subscribers, replacing them with a rate-limit reset usable through October 22. External evaluators including METR tested the model, which achieved the company's strongest internal safety audit results to date.
Key Capabilities:

• Fable-level performance: Matches Claude Fable 5.1 on most tasks at 40% lower cost than Opus 5.
• No more usage caps: Five-hour limits removed for Pro, Max, Team, and Enterprise plans.
• Stronger safety: Achieved Anthropic's best internal behavioral audit score to date.