Phala

Phala Confidential AI: 168.69 billion tokens over the past 24 hours. Top models: GLM-5.3 42.87% deepseek-v4.1-flash 19.24% GLM-5.3 Flash 12.00% https://lnkd.in/gTXc-TB7 TEE-backed AI infrastructure.

Phala Confidential AI: 55.18 billion tokens over the past 24 hours. Top models: GLM-5.3 Flash 49.10% GLM-5.3 14.71% Kimi K3 12.62% https://lnkd.in/gEVMa8Fn TEE-backed AI infrastructure.

Phala Confidential AI: 78.09 billion tokens over the past 24 hours. Top models: GLM-5.3 43.54% GLM-5.3 Flash 26.96% DeepSeek V4 Flash 11.46% https://lnkd.in/gTXc-TB7 TEE-backed AI infrastructure.

Qwen3 8B is live on Phala—served inside a TDX-attested GPU TEE. Private, verifiable inference for coding, long-horizon agents, reasoning. https://lnkd.in/gkUv9uSu 41K context. 8.2B parameter causal language model from the Qwen3 series. Reasoning, tool use, structured outputs. $0.11/M input · $0.45/M output. https://lnkd.in/e4yu_p_C

Phala Confidential AI: 50.72 billion billed input + output tokens over the past 24 hours. Top models: 1) z-ai/glm-5.3-flash 54.28% 2) deepseek/deepseek-v4-flash 17.77% 3) z-ai/glm-5.3 7.01% https://lnkd.in/gEVMa8Fn TEE-backed AI infrastructure.

Phala Confidential AI: 38.86 billion billed input + output tokens over the past 24 hours. Top models: 1. DeepSeek V4 Flash — 28.59% 2. GLM-5.2 — 13.61% 3. GLM-5.3 — 9.53% Explore #1: https://lnkd.in/eN_N7C9Y TEE-backed AI infrastructure.

ViMax transforms ideas, novels, or scripts into multi-scene video plans and final generated clips. The Phala template uses a source_verifier runtime. https://lnkd.in/d4Tre83r By default, this template does not run video generation. Its /demo constructs upstream Pydantic models for a scene, shot brief, shot description, camera, and event. https://lnkd.in/dre44EFt

39.81 billion tokens powered Confidential AI on Phala over the past 24 hours. Top models: 1. DeepSeek V4 Flash — 29.90% 2. GLM-5.2 — 17.19% 3. Kimi K3 — 14.98% Explore #1: https://lnkd.in/eN_N7C9Y TEE-backed AI infrastructure.

Phala Confidential AI network token volume: 42.98 billion billed input + output tokens over the past 24 hours. Top models 1. Kimi K3 — 35.48% 2. DeepSeek V4 Flash — 22.56% 3. Qwen3.5 397B A17B — 8.50% https://lnkd.in/ej-R4w6x TEE-backed AI infrastructure.

GLM-5.3 from @Zai_org is live on Phala—served inside a TDX-attested GPU TEE. Built for complex software engineering and long-horizon agents, with 1M context, tool use, and structured outputs. $1.40/M input · $4.40/M output. https://lnkd.in/gTXc-TB7 We built GLM-5.3-W4AFP8 directly from the BF16 master. MoE expert weights use group-128 INT4 with AWQ; activations and non-expert layers use FP8. Calibration uses coding-agent traces from SWE-chat, aligned with long-horizon agent workloads. The result: roughly 2× KV-cache capacity vs FP8 at matched throughput, with the full 1M context on 8×H200. EAGLE/MTP speculative decoding stays intact, with an average accept length of ~2.93. Benchmark checks: 91.92 GPQA-Diamond (182/198), 82.2 on a 45-item BFCL subset, and 3/3 NIAH retrieval at ~930k-token prompts. Weights, serving config, and per-item evaluation artifacts: https://lnkd.in/gBPTnHXK