DeepSeek published the models, the harness and the 88-page paper under it
Then kept every researcher off camera This one talked for 59 minutes, and he explains why your agent gets dumber the longer you run it 19:04 - agents already past the self-improvement line 20:11 - why an
A brand-new coding tool hit 90K GitHub stars in 2 days.
We had to test it against the tool we use every single day. 👀 DeepSeek Harness dropped on August 13 open-source, free, and built so you can swap out the model, tools, and memory like Lego bricks. So we ran the exact
5 cents. That's the extra cost for DeepSeek V4 Pro to inspect and rebuild Linear's homepage with ego lite, seven minutes faster than V4 Flash.
Same harness. Same prompt. Grok 4.6 took 33 minutes and $5.06, more than twice as long as Pro while costing roughly 10x as much. Three
Qwen 3.8 27b
DeepSeek V4 Flash 0731 GPT 5.6 Luna Most inexpensive models fail these tests.
You have reached the end of the archive
All of deepseek