Daily Brief

DeepSeek has closed a $10.29 billion financing round, with founder Liang Wenfeng publicly committing to prioritize open-source AI development rather than near-term revenue generation. This is one of the largest single AI funding rounds outside of OpenAI and Anthropic, and it's explicitly being directed away from commercialization.

DeepSeek has converted its promotional 75% API pricing discount into a permanent price structure, meaning its models now cost roughly one-quarter of comparable OpenAI and Anthropic endpoints on an ongoing basis. The move comes concurrent with its $10.29B raise, ruling out financial distress as the driver.

Google's Gemini 3.5 Flash — a smaller, faster model — has taken the top spot on the APEX-Agents-AA benchmark, outperforming models in the size tier above it on agent task execution. This is the second major efficiency win for a sub-frontier model on an agentic eval in recent weeks.

Gartner placed OpenAI in the Leaders quadrant of its 2026 Magic Quadrant for Enterprise AI Coding Agents, citing Codex specifically for innovation and deployment scale. The Virgin Atlantic case study — near-total test coverage and zero P1 defects shipped on deadline — was published simultaneously as supporting evidence.

Anthropic published an initial update from Project Glasswing, an internal research initiative. Details from the announcement suggest this is an active safety or interpretability research program now surfacing external findings.

A study using 82 Nature-submitted papers found GPT-5.2 performing at the level of top human peer reviewers, with identified but limited failure modes. This is not a PR release — it's an academic study against one of the highest-prestige publication venues in science.