San Francisco lab releases 399-billion parameter reasoning model under Apache 2.0 license, filling the void left by Meta's retreat from open weights.
Z.ai's GLM-5.1 achieves breakthrough performance on SWE-Bench Pro, running autonomously for 8 hours on single tasks under MIT license.
Tsinghua University and Z.ai researchers release IndexCache, a sparse attention optimizer that cuts 75% of redundant computation in DeepSeek-based LLMs, delivering 1.82× faster inference at 200K token context lengths.
Researchers at Tsinghua University have developed IndexCache, a new technique that achieves 1.82x faster inference for long-context AI models by eliminating redundant sparse attention computations.