
AI·3 min read
DeepSeek V4 Pro at 91 or 99, depending on which harness runs the test
Same model, same frozen task, a score that lands anywhere from 91 to 99. What the headlines read as an eight point jump is the spread of the benchmark itself.
Tags
2 stories found

Same model, same frozen task, a score that lands anywhere from 91 to 99. What the headlines read as an eight point jump is the spread of the benchmark itself.

DeepSeek released a developer preview of DeepSeek Harness v0.1. Its Cordis design makes the model, tools, sessions, execution loop, sandbox and web UI all replaceable plugins.