AI news for Tuesday, October 13, 2026: Meta ships Muse Wonder, Anthropic locks down Claude's reasoning

A frontier launch with no numbers, and a lab tightening the rules you build under. Test before you trust either.

Models Agents3 min
Meta ships Muse Wonder with no benchmarks or prices
Meta released Muse Wonder and Muse Wonder Full on Oct. 12, calling them its most capable models, available now through the Meta Model API and Muse Code. The post gives no scores, no prices and no safety report yet, so the frontier claim is Meta's until you test it.
Why it matters If you pick models for real work, Meta just became a vendor worth a test run. Put Muse Wonder on one of your own agent tasks this week. The launch post gives you no numbers, so yours will be the only ones you have.
SRCMeta
Safety Models3 min
Anthropic says it cut Claude distillation by up to 99%
Anthropic says new defenses cut the Claude reasoning data distillers harvest by about 90 to 99%, and it has banned tens of thousands of accounts. The fine print reaches honest developers too: new API accounts can't edit Claude's prior context, and blocked requests are billed.
Why it matters If you opened a Claude API account on or after Aug. 31, check whether your code edits earlier turns, because Anthropic says preserved thinking has required some developers to update their integrations. And budget for the rare false block: Anthropic charges for requests its safeguards stop before Claude answers.
SRCAnthropic
Hardware Business3 min
Microsoft makes Maia its main AI inference chip
Microsoft says Maia 200 is in production and Maia 300 starts multi-gigawatt deployments in 2027, making its own chip the primary inference system for its products. GitHub Copilot and Foundry move to Maia by year end, so your requests will soon run on Microsoft silicon.
Why it matters If you build on Microsoft Foundry or code with GitHub Copilot, the hardware under your requests changes by year end, and Microsoft is betting it serves tokens cheaper. Keep a latency and cost baseline from before the switch so you can tell whether that bet reaches you.
SRCMicrosoft
Business Products3 min
Character.AI files for Chapter 11, Disney bids $20M
Character.AI filed for Chapter 11 in Delaware on Oct. 12 to sell nearly all its assets, with Disney as the $20 million opening bidder for a license to some of its IP. One of the best known AI companion apps is now selling itself in pieces, and whether it stays online depends on who bids.
Why it matters If you use Character.AI, the app keeps running for now, but the company says it will be wound down if no buyer wants to operate it. Save what you care about while it is up. If you build a consumer chat app, note what the first bidder paid for: an IP license and staff it already hired.
SRCStretto (Character.AI case site)
Creative Business3 min
Epic orders an AI-first Disney game, Bloomberg reports
Epic Games delayed its Disney game and told the team to go "AI first," with "no more hand-written code," Bloomberg reported Oct. 12, citing unnamed sources. If the order holds, a flagship game becomes a public test of full AI production, and the result will be in stores for anyone to judge.
Why it matters If you lead a creative or engineering team, this is the mandate you may hear next: no hand-written code, art from AI workflows. Before anyone copies it, decide what "AI first" means in your shop, who reviews the output and who owns the bugs.
SRCEngadget (relaying Bloomberg)

Prompt of the day

Build a fair test for any new AI model on your own work

› I want to decide whether a newly released AI model is worth switching to for my work. Its maker has published no benchmark scores I can use, so I need my own test.

My work: [describe what you use AI for, e.g. reviewing contracts, writing support replies, building a web app]
The model I use now: [current model]

Please:
1. Propose 5 test tasks drawn from my real work, from routine to hard, including at least one multi-step task where the model has to recover from its own mistake.
2. For each task, write the exact prompt I should paste into both models, word for word, so the comparison is fair.
3. Give me a scoring sheet with 3 to 5 criteria per task (accuracy, follow-through, how many corrections I had to make, time to a usable result), each scored 1 to 5, with what a 1 and a 5 look like.
4. List what I should not paste into a new model before its safety report and data terms are public, given my kind of work.
5. Tell me how to decide: what score gap would justify switching, and what would make me stay.

Ask me any questions you need first, then build the test.

Open in Magai

Works in any AI chat. In Magai, run it against several models at once.

Get every edition

This one arrived in inboxes at 6am ET.

Meta ships Muse Wonder with no benchmarks or prices

Anthropic says it cut Claude distillation by up to 99%

Microsoft makes Maia its main AI inference chip

Character.AI files for Chapter 11, Disney bids $20M

Epic orders an AI-first Disney game, Bloomberg reports