Model Releases
DeepSeek V3.2: The 671B MoE Open Source King
DeepSeek AI releases V3.2 on Feb 12, 2026. A 671B MoE model with 1M context window, MIT licensed, and open weights on HuggingFace.
6 min read
Latest news, tutorials, and insights about AI
DeepSeek AI releases V3.2 on Feb 12, 2026. A 671B MoE model with 1M context window, MIT licensed, and open weights on HuggingFace.
Zhipu AI launches GLM-5, China's first public frontier model, optimized for complex engineering and agentic workflows.
OpenBMB launches MiniCPM-o 4.5, a 9B parameter multimodal LLM matching Gemini 2.5 Flash performance with on-device capabilities.
OpenAI releases GPT-5.3-Codex on February 5, 2026, introducing a 25% faster, 1-million-token context window agentic coding model optimized for full software engineering workflows.
Anthropic releases Claude Opus 4.6 on February 5, 2026, marking a historic leap in agentic planning with parallel subtask execution and record-breaking terminal performance.
StepFun unveils Step-3.5-Flash, a 196B MoE model delivering frontier reasoning at 350 tokens per second with open weights.
Arcee AI releases Trinity Large, a 400B parameter MoE model with Apache 2.0 licensing, challenging proprietary giants with sovereign infrastructure.
Alibaba Cloud unleashes Qwen3-Max-Thinking, a top-tier reasoning model that rivals frontier competitors with adaptive tool use and code execution.
Moonshot AI releases Kimi K2, a 1T parameter open model ranking #1 on LMSYS. See pricing, benchmarks, and specs for developers.