This brief covers AI news from 2026-11-18 UTC.
Two frontier labs shipped model families on the same day, and the fight is now about cost per task rather than raw scores.
OpenAI ships GPT-6.2 Astra and Sol with 20% cheaper tokens
What happened: OpenAI released GPT-6.2 Astra for hard professional work and GPT-6.2 Sol for everyday tasks. Astra’s API input and output prices are 20% lower than GPT-6 Astra’s, with cached input at $0.40 per million tokens. OpenAI says Astra beats Claude Fable 5.2 by 5.5 points on Agents’ Last Exam at 5% of the cost.
Why it matters: Cheaper frontier tokens and a 60% cache discount change the math on long agent runs, so recheck your per-task budgets against the new prices.
Microsoft launches seven MAI models at Ignite, led by cheap coding
What happened: Microsoft AI released seven models, including MAI Code 2, MAI Code 2 Flash, MAI Image 2.7, MAI Realtime 1 and MAI Cyber 2 Flash. Microsoft says MAI Code 2 hits 98% of GPT-6.1 Sol’s coding quality at 40% of the price, and that Maia 200, its second-generation accelerator, is now in production.
Why it matters: A second price-performance option for routine coding and high-volume edits gives teams leverage when routing work between a frontier model and a cheap one.
AMD closes its $8.2 billion World Labs acquisition
What happened: AMD completed its purchase of World Labs, the world-model startup founded by Fei-Fei Li in 2024, paying roughly $8.2 billion entirely in stock. Li becomes chief scientist at AMD and reports to CEO Lisa Su. The deal was announced in late September.
Why it matters: World Labs’ Marble models already train and run on AMD GPUs, so the acquisition ties 3D world-model tooling more tightly to AMD hardware.
Google open-sources ABFS to speed up giant Git checkouts
What happened: Google released ABFS, a content-addressable build file service, on GitHub. On Android it cut source checkout from about 36 minutes to under 8 seconds and full builds from about 80 minutes to just over 2 minutes, and it enables parallel agentic workflows across workspaces.
Why it matters: If your agents spend their first minutes syncing a monorepo, a lazy-loading filesystem layer is the difference between fast iteration and idle waiting.
Source: Google Developers Blog
GitHub details the Agent Host Protocol for cross-device sessions
What happened: GitHub published how it built the open Agent Host Protocol, which keeps a Copilot agent session alive across VS Code, the Copilot app, mobile and the cloud. It was built first in VS Code and adopted across GitHub, replacing per-client integrations that died with the editor window.
Why it matters: An open protocol for session continuity means you can build your own client for a long-running agent instead of writing a one-off integration per app.
Workday moves to buy Honeydew for governed agent definitions
What happened: Workday announced its intent to acquire Honeydew, a semantic layer for AI and business intelligence. Honeydew’s semantic compiler turns business definitions into governed queries so agents and BI tools return consistent answers within each user’s permissions.
Why it matters: If your agents query enterprise data, the semantic layer is where inconsistent definitions get fixed, and this deal pushes that layer into an HR and finance platform.
Model ML raises $250 million at a $1.5 billion valuation
What happened: Model ML raised a $250 million Series B co-led by General Atlantic and QED Investors at a $1.5 billion valuation, bringing total funding to $350 million. The company says its annual recurring revenue grew 12x year on year in the third quarter of 2026.
Why it matters: Money keeps flowing to vertical agents for finance work, a segment where the product is judged on whether the analysis is already done before the user asks.