Note: Adapted from the official X Phala Network announcement. Check here: https://x.com/phalanetwork/status/2046629390889754758
If you’ve been building on Phala Cloud, there’s a meaningful update worth knowing about. Since late February, a new set of models has been added to the platform, and the range covers a lot of ground, from multimodal reasoning to heavy-duty coding agents to large-scale inference workloads. Whether you’re a solo developer, a team shipping a product, or an institution that needs to keep inference data locked down, there’s likely something in here relevant to what you’re building.
more capable models are now available, and they all run the same way everything on Phala Cloud runs, inside a TEE-secured environment that keeps your data private by default.
What Models Were Added
Six notable additions landed since late February.
1️⃣ MoonshotAI’s Kimi K2.6 brings multimodal capability and is well suited for multi-agent workflows and coding-driven UI generation. Check here: https://www.redpill.ai/models/moonshotai/kimi-k2.6
2️⃣ Qwen3 Coder Next from Qwen is built specifically for coding agents and local development, using a sparse architecture that keeps things efficient. Check here: https://www.redpill.ai/models/qwen/qwen3-coder-next
3️⃣ GLM 5.1 from Z.ai is a strong pick for complex, long-running engineering tasks where you need a model that can hold context and stay useful across a full workflow. Check here: https://www.redpill.ai/models/z-ai/glm-5.1
4️⃣ On the larger end, Xiaomi’s MiMo-V2-Flash is a 309 billion parameter model using a Mixture-of-Experts setup, which makes it practical for demanding workloads without burning through compute unnecessarily. Check here: https://www.redpill.ai/models/xiaomi/mimo-v2-flash
5️⃣ Qwen3.5 397B A17B is a vision-language model combining two different architectural approaches to deliver solid performance across reasoning, coding, and interface interactions. Check here: https://www.redpill.ai/models/qwen/qwen3.5-397b-a17b
6️⃣ Qwen3.5–27B rounds things out as a lighter, faster option that still holds its own against much larger models in most practical tasks. Check here: https://www.redpill.ai/models/qwen/qwen3.5-27b
Why the Infrastructure Behind These Models Matters
Adding new models to a platform is straightforward. What’s less common is running all of them inside hardware-level private infrastructure. Every model on Phala Cloud runs inside a Trusted Execution Environment, secured by Intel TDX and NVIDIA Confidential Computing. That means the computation itself is isolated and verifiable, not just the data in transit.
For individual developers, this means you can build and test against powerful models without worrying about your prompts or outputs being exposed. For teams and companies, it means your inference workload doesn’t become a liability. For institutions handling regulated or sensitive data, it means you can actually use these models in production without a long conversation with your compliance team first.
This is the core of what Phala Cloud is doing differently. It’s not just a hosted model catalog. It’s a place where the privacy guarantee is built into the hardware layer, not bolted on afterward.
How to Get Started
The full catalog currently includes over 200 models, and the six highlighted above are a good starting point depending on what you’re building. If you’re working on a coding agent, Qwen3 Coder Next is worth testing first. If you need multimodal capability or multi-agent coordination, Kimi K2.6 is the natural starting point. For large-scale or institution-level workloads, MiMo-V2-Flash gives you serious capacity with the same privacy guarantees as everything else on the platform.
You can explore the full model list and start building at Phala Cloud today. The infrastructure is already running. You just need to bring the idea.
Check the full list here: https://www.redpill.ai/models
Top comments (0)