Open-weight and domain-specific model competition expands

Tencent’s reported Hy3/Hy4 coding MoE releases, DeepSeek and Qwen continue to pressure frontier-model TCO, while ChemELLM targets chemistry, engineering and plant operations. IBM Granite Speech 5.0 adds an enterprise speech-model option optimized for throughput and local deployment, though language coverage and vendor-reported benchmarks limit the evidence.