DeepSeek v4, and the end of the OpenAI/Microsoft AGI clause
16 hours ago
- #LLM Development
- #OpenAI AGI
- #AI Models
- DeepSeek released V4 series models, including DeepSeek-V4-Pro and DeepSeek-V4-Flash, which are large-scale Mixture of Experts models with 1 million token context and competitive performance.
- DeepSeek models are notably cheaper than other frontier models, with Flash priced at $0.14/million tokens input and Pro at $1.74/million tokens input, due to efficiency improvements.
- OpenAI and Microsoft's AGI clause, which nullified Microsoft's IP rights upon AGI achievement, has ended, with Microsoft now having a non-exclusive license through 2032.
- LLM 0.32a0 introduces a major refactor to handle diverse input/output types, including messages as sequences and streaming of mixed content parts like text and tool calls.
- The newsletter includes various links and updates, such as new pip 26.1 features, Microsoft's VibeVoice speech-to-text model, and discussions on AI automation's public reception.
- Zig project enforces a strict ban on LLM-assisted contributions, emphasizing contributor growth over code, while other projects like Bun use AI heavily.