X

Honoured to participate in the German Federal Cabinet Retreat this week to discuss key areas of focus for Germany, including how it can empower its industries and ensure AI competitiveness. My sincere thanks to Chancellor @_FriedrichMerz for the invitation. https://t.co/3dwICaUycg

Introducing GLM-5.3-Flash - Leading capabilities at a highly competitive price - Natively multimodal with a 1M-token context window - A 320B-A18B model released under the MIT License - Previously previewed as Ox Alpha, running entirely on Chinese AI chips Blog: https://t.co/tzOmB7gdZP Available now across all official platforms: Weights: https://t.co/9LRMahY9Wa API: https://t.co/VcaQnzYmS9 Coding Plan: https://t.co/Nk8Y98HNhU ZCode: https://t.co/Peepqv4XSx Chat: https://t.co/WCqWT0qCQb AutoClaw: https://t.co/aGEG5HqTTb

GLM-5.3 Flash ("Ox Alpha") official: Benchmarks attached. This looks exceptional for its size! GLM-5.3-Flash might be one of the most impressive efficiency releases yet. It is a 320B MoE with only 18B parameters active per token, yet Zai reports: - 84.3 on Terminal-Bench 2.1, nearly matching Claude Opus 4.8 at 85.0 - 63.4 on DeepSWE, ahead of Opus 4.8 and DeepSeek V4 Vision Exp - 48.8 on AutomationBench, ahead of Opus 4.8 and GPT-5.6 Terra - The highest GDPval-AA v2 score in its comparisonIt also beats the much larger GLM-5.2 across all six reported benchmarks while costing one-tenth as much to serve. Open weights, MIT licensed, natively multimodal, 1M context. Important caveat: 18B active parameters does not make it a normal local 18B model. All 320B weights still need to be stored. But in terms of intelligence per active parameter, this looks exceptional!

The upcoming Ox Alpha is GLM-5.3 Flash (as expected): 320b total parameters, 18b active. Outperforming GLM-5.2 at 1/10th of its price and approaching Opus 4.8 on coding and agentic benchmarks. Big things incoming! https://t.co/xZZmOD1Ghu https://t.co/H2fZ9idgVa

Your creations can earn. PixVerse Earn is now live. A new way for AI video creators to turn PixVerse-made videos into earning opportunities through official campaigns. Create with PixVerse. Post anywhere. Earn for real. Follow + RT + Reply = 150 Creds in Dms (72h only) https://t.co/GkHtd6LHIM

⚡Meet Qwen3.8-Flash, a multimodal MoE and an early preview of the Qwen4 architecture, now open-weight! The production version Qwen3.8-Flash will be available soon via QwenCloud API at just $ 0.16/1M input tokens and $ 0.47/1M output tokens. 125B parameters + 51B N-gram embeddings, with just 6B activated per token. Unmatched cost-efficiency. What's new: 🥳 - Next architecture: GDN + QSA hybrid attention, Gated Residual, N-gram Embedding & Muon optimizer, serving as a precursor to the architecture used in Qwen4. - Dramatically lower training and inference costs: trained at just 1/9 the cost of Qwen3.7-Plus, while outperforming it across the board with especially strong gains in coding and office tasks. - Strong performance: scoring 58.7 on DeepSWE 1.1, 62.5 on SWE-bench Pro, 73.9 on CoWorkBench, 84.5 on AndroidWorld, and 95.7 on MathVision (with CI). - 262K native context, extensible to 1M with YaRN. We’re also releasing the weights for Qwen3.8-Flash-Next, giving the community an early look at the new architecture we’re exploring for Qwen4.🚀 We can't wait to see what you build with Qwen3.8-Flash!👀👇 - Blog: https://t.co/M5hYypFLgJ - Technical Report: https://t.co/IF0gObIkQO - Hugging Face: https://t.co/6ow8QVAABt - ModelScope: https://t.co/tDOn2jNuFG

Model Architecture Four core upgrades for maximum capability, efficiency, capacity, and stability: - Attention: GDN + QSA Hybrid. Gated DeltaNet (GDN) compresses history. Qwen Sparse Attention (QSA) uses a lightweight indexer for micro-block context selection. Lower the cost of attention on long sequences. - Residual: Gated Residual (GR) widens the residual stream to 4 branches with a dynamic read and write gating, strengthening cross-layer information flow and significantly improving training stability. - Embedding: N-gram Embedding uses local context lookups to expand model capacity at minimal compute cost, while keeping the embedding table in host memory with asynchronous prefetching. - Optimization: Muon optimizer. Refines Muon through improved orthogonalization, smarter parameter assignment between Muon and AdamW, and fused-parameter splitting, with scaling laws refitted for the new architecture.