DeepSeek has released its V4 models, including V4-Pro and V4-Flash, both featuring a 1 million token context window and released under the MIT license. These models achieve performance on par with Claude Opus 4.6 on public agent benchmarks and surpass GPT-5.4 on coding challenges, positioning them as strong open-weight alternatives for AI agents. Key improvements include a new XML-based schema for tool calls that reduces parsing errors and enhanced reasoning persistence across multiple tool interactions. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Sets new SOTA on agentic benchmarks for open-weight models, offering a cost-effective alternative to proprietary models for tool-driven agents.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]