Open-model release wave (DeepSeek V4, Qwen3.7-Max, Stable Audio, DR Tulu)
Key Questions
What improvements does DeepSeek V4 offer over prior versions?
DeepSeek V4 provides a 10x reduction in inference costs under its MIT release. The Pro tier is now priced at one-quarter of the original rate, enabling more agentic applications.
What new Qwen models were released and what are their strengths?
Qwen3.7-Max targets the agent frontier while Qwen3-Coder-Next focuses on coding agents. Both are part of Alibaba's full-stack AI upgrades for the agentic era.
What is DR Tulu and its intended use case?
DR Tulu 8B is a fully released open model designed for long-form deep research tasks. It supports training open models on extended research workflows.
How does Cohere Command A+ compare in licensing and performance?
Command A+ is released under Apache 2.0 with fully open weights and achieves 281 tokens per second. It offers strong visual reasoning and enterprise transparency.
What features does Stable Audio 3.0 provide?
Stable Audio 3.0 is an open-weight model family supporting variable-length audio generation. It is built for artistic experimentation by Stability AI.
How do GLM-5.1 routing benchmarks impact costs?
GLM-5.1 open-weight routing shows 31% cost reductions compared to closed models like Claude Opus 4.7. Benchmarks were run on platforms such as TrueFoundry AI Gateway.
Where can users access these new open models?
Models are now available on Vercel and Microsoft Foundry. DeepSeek V4 and related releases are hosted on GitHub with MIT licensing for broad adoption.
What other open models were highlighted in recent releases?
Kimi K2.6 outperforms several closed models on vibe coding tasks. Cartesia's Sonic-3.5 leads speech benchmarks while Gemma family updates continue from Google.
DeepSeek V4 MIT release with 10x inference cost reduction (Pro now 1/4 price). Qwen3.7-Max agent frontier; Qwen3-Coder-Next for coding agents; DR Tulu 8B full release; Cohere Command A+ Apache 2.0 (281 tok/s); Stable Audio 3.0 open-weight with variable-length. GLM-5.1 open-weight routing benchmarks show 31% cost cuts. Now on Vercel and Microsoft Foundry.