# Meta Launches Muse Wonder Models; Anthropic Enhances Security Meta's new AI models, Muse Wonder and Muse Wonder Full, aim to enhance coding and long-running tasks. Muse Wonder is available now at $4 per million input… **Published:** 2026-10-13 · **Author:** Yash Thakker **Source:** https://explainx.ai/newsletter/meta-launches-muse-wonder-models-anthropic-enhances-security --- > **In short:** Meta's new AI models, Muse Wonder and Muse Wonder Full, aim to enhance coding and long-running tasks. Muse Wonder is available now at $4 per million input tokens, while Muse Wonder Full is exclusive to subscribers. ## Meta Launches Muse Wonder and Muse Wonder Full Models ### Meta Muse Wonder and Wonder Full: Pricing, Benchmarks and How to Get Access Meta has introduced two new AI models, Muse Wonder and Muse Wonder Full, aimed at coding and long-running tasks. Muse Wonder is priced at $4 per million input tokens and $20 per million output tokens, while Muse Wonder Full is available only to Muse Code subscribers. Meta claims competitive performance against leading models, but details on specific capabilities and pricing for Wonder Full remain unclear. **What you can do:** Consider exploring Muse Wonder if you're looking for advanced coding capabilities, but be mindful of the costs involved. [Source](https://explainx.ai/blog/meta-muse-wonder-full-launch-pricing-benchmarks-muse-code-2026) ## AI Developments: Distillation Defenses, On-Device Models, and Self-Replication ### Anthropic Says New Distillation Defenses Cut Stolen Reasoning Data by 90 to 99 Percent Anthropic has introduced three defenses that significantly reduce the data distillers can extract from its Claude models, claiming a 90 to 99 percent drop in harvested data. This includes a new classifier with a false-positive rate below 0.1 percent. [Source](https://explainx.ai/blog/anthropic-distillation-defenses-preserved-thinking-classifier-90-99-percent-2026) ### MiniCPM-V 4.7: A 35B MoE Vision Model That Runs on an iPhone OpenBMB launched MiniCPM-V 4.7, a 35-billion-parameter vision model that reportedly decodes at over 13 tokens per second on an iPhone 17 Pro Max. This model uses a mixture-of-experts approach for efficient processing. [Source](https://explainx.ai/blog/minicpm-v-4-7-35b-a3b-on-device-moe-iphone-lookahead-expert-scheduling-2026) ### Anthropic: Small AI Models May Soon Self-Replicate on Smart TV Boxes Anthropic's research suggests that small AI models could soon replicate across inexpensive IoT devices, like smart TVs. Their tests showed a model successfully replicating on various hardware, raising concerns about future security risks. [Source](https://explainx.ai/blog/anthropic-self-replicating-ai-iot-smart-tv-botnet-frontier-red-team-2026) ## Understanding AnyPS5: Running PS5 Games on PC ### How AnyPS5 Runs PS5 Games Natively on PC Without an Emulator AnyPS5 converts PS5 executables into native PC applications without emulation. Currently, only one 2D game, Dreaming Sarah, is confirmed playable, with the project still in development. **What you can do:** If you're interested in gaming on PC, keep an eye on AnyPS5's progress. While it currently supports only limited titles, its approach could expand compatibility in the future. [Source](https://explainx.ai/blog/anyps5-ps5-games-native-pc-no-emulator-how-relinking-works-2026) ## Workshop on AI Decision-Making with Jev ### Build with Jev: The AI That Decides, Not Writes Join a 2-day live workshop on Jev, TypeSafe's System One model. Learn about decision-making with AI, including typed choices, score and Noul decisions with calibrated confidence. Starts October 20, 2026, for about 2 hours each day. [Source](https://explainx.ai/workshops/jev) ## Reader poll What do you think about the potential for small AI models to self-replicate on IoT devices? - It's a serious concern for security. - It could lead to innovative applications. - I'm not sure how to feel about it. - I don't think it will happen. ## Sources - [Meta Muse Wonder and Wonder Full: Pricing, Benchmarks and How to Get Access](https://explainx.ai/blog/meta-muse-wonder-full-launch-pricing-benchmarks-muse-code-2026) - [Anthropic Says New Distillation Defenses Cut Stolen Reasoning Data by 90 to 99 Percent](https://explainx.ai/blog/anthropic-distillation-defenses-preserved-thinking-classifier-90-99-percent-2026) - [MiniCPM-V 4.7: A 35B MoE Vision Model That Runs on an iPhone](https://explainx.ai/blog/minicpm-v-4-7-35b-a3b-on-device-moe-iphone-lookahead-expert-scheduling-2026) - [Anthropic: Small AI Models May Soon Self-Replicate on Smart TV Boxes](https://explainx.ai/blog/anthropic-self-replicating-ai-iot-smart-tv-botnet-frontier-red-team-2026) - [How AnyPS5 Runs PS5 Games Natively on PC Without an Emulator](https://explainx.ai/blog/anyps5-ps5-games-native-pc-no-emulator-how-relinking-works-2026) - [Build with Jev: The AI That Decides, Not Writes](https://explainx.ai/workshops/jev) --- From [explainx.ai](https://explainx.ai/newsletter). Read online, comment and subscribe: https://explainx.ai/newsletter/meta-launches-muse-wonder-models-anthropic-enhances-security