Intel / post

DeepSeek V4 Pro 0813 posts large agent benchmark jumps over its preview

Source: x.com
Date: Published Aug 12, 2026

Context

DeepSeek's V4 Pro 0813 release reported a substantial improvement in completing computer tasks through an . OpenRouter's announcement gives a Terminal-Bench 2.1 score of 87.9, up 15.8 points from V4 Pro Preview, and DeepSeek's release log lists the same score. Terminal-Bench tests agents on tasks performed in a terminal, each with test cases that check the agent's solution. The comparison therefore concerns executing tasks through software tools, rather than simply producing convincing written answers. It remains a developer-reported result.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.

Read the source x.com