[AINews] AMD buys World Labs for $8.2B, as Atlas solves sparse reconstruction problem for robotics, design and more

What it claims

Two items. First, AMD is acquiring World Labs for $8.2B. The price is known because AMD is public; the official post does not state it. World Labs (founded 2024) describes Atlas as an omni-architecture model trained from scratch. Given 2D images, it predicts new camera views, and World Labs says this "essentially solved" sparse reconstruction by combining generative models with multiview geometry. World Labs has also acquired SceniX and is using it to build robotics simulation.

Second, the issue's top story is Claude Sonnet 5.5, the second model in the Claude 5.5 family. It shipped one week after Opus 5.5 and the day before OpenAI DevDay. Anthropic calls it "a clear upgrade over Sonnet 5." Haiku 5.5 is due "in the coming weeks" to complete the family.

Evidence

• Sonnet 5.5 speed and cost: Anthropic claims it runs more than 30% faster than Sonnet 5 and costs up to 30% less "for most work." The page gives no benchmark breakdown behind these figures.
• Independent evals: early results place Sonnet 5.5 at or near Opus 5.5 on several leaderboards. The page does not name the leaderboards or give scores.
• Atlas: World Labs says Atlas outperforms state-of-the-art results, including specialized models. No benchmark, metric or baseline is given.

How to use it

• Where it runs: Sonnet 5.5 shipped day one on the Claude Platform and Claude Code. It also now powers the claude.ai free tier. ChatGPT's free tier is still GPT-5.6 Luna, which Simon Willison calls "a lot less capable."
• Usage reset: a usage reset came with the launch and is valid until Oct 22.
• Choosing a model: Anthropic published a build guide that covers:
• when to pick Sonnet 5.5 vs. Opus 5.5
• migrating from Sonnet 5
• tuning effort
• Intended scope: Anthropic positions Sonnet 5.5 for "well-scoped everyday tasks like fixing bugs and quickly iterating on features." Longer or open-ended work points toward Opus 5.5.
• Preserved thinking across accounts: this behaviour changes multi-account workflows.
• Reasoning traces stay in the org that generated them.
• If a session moves to another account, Claude rereads the session and regenerates its thinking instead of carrying the traces over.
• Expect re-generation cost and possibly different reasoning when you hand sessions between orgs or accounts.

Limits

• Atlas claims are unverified here. "Solved" and "outperforming state of the art" come from World Labs' own reflection post, with no numbers, datasets or access details. The applications named (robotics RL environments, scene generation for therapy and entertainment, reconstruction for real estate, design and construction) are areas of interest, not reported results.
• No Atlas availability details. The page does not say whether Atlas is accessible, or how, after the acquisition.
• Sonnet 5.5 figures are vendor claims. The speed and cost numbers come from Anthropic's announcement, and the independent eval claims carry no figures.
• The fetched text is truncated. It cuts off at the third-party availability list for Sonnet 5.5, so providers outside Anthropic's own platforms are not established here.