Open-weight model surge
Key Questions
Which open-weight models are surging in popularity?
Qwen3.5, Gemma 4, DeepSeek V4 Pro, Kimi K2.7-Code, and GLM-5.2 are among the leading open-weight models gaining traction.
How does GLM-5.2 impact frontier lab margins?
GLM-5.2 achieves Opus-level performance at lower cost on CursorBench, driving down margins for frontier labs according to Nathan Lambert.
What is the status of DeepSeek funding and valuation?
DeepSeek confirmed $7.4B in funding at a $50B valuation while introducing vision capabilities.
What recent release did NVIDIA make for open models?
NVIDIA released an NVFP4-quantized version of MiniMax-M3, a 428B parameter model with 1M context length.
How are open-weight models being integrated commercially?
Tencent integrated DeepSeek V4 into WeCom, and Llama.cpp received official branding support.
Open-weight models surging: Qwen3.5, Gemma 4, DeepSeek V4 Pro, Kimi K2.7-Code, GLM-5.2. GLM-5.2 (753B MoE, MIT license) beats GPT-5.5 on SWE-bench Pro at 1/6 cost, with NVIDIA optimizing for Blackwell. Ornith-1.0 (397B) learns own RL scaffold and beats Opus 4.7. Llama.cpp gets official branding. Codex now supports open-source models. DeepSeek introduces vision, $7.4B funding. NVIDIA releases NVFP4-quantized MiniMax-M3 (428B, 1M context). Tencent integrates DeepSeek V4 into WeCom. The Unbearable Cheapness of Open Weight Models validates economic shift.