148 times cheaper. 84% of the quality.
Three models got the exact same prompt: build a Flappy Bird game. Kimi K3 and GLM 5.2 drew the better-looking output (9.5/10 and 9/10), DeepSeek V4 Pro came in a bit lower (8/10). Alongside the release, DeepSeek announced a price hike. "We
Visited Hangzhou’s “Six Little Dragons 🐉” tech center.
DeepSeek, Unitree, Game Science, BrainCo, DEEP Robotics and Manycore all come from Hangzhou! Big shoutout to the staff for turning on all the robot demos just for me because they are usually only for groups ✌️
Nobody ever shows you the log.
Here it is: 687 steps of a local Qwen3.8 27B building a first person shooter in DeepSeek Harness, scrolled end to end. It takes a while. 87.9M tokens went in, 822k came out. 5 hours 11 minutes of model time, 58 minutes of tool calls. All of it on
Ditch the $200-a-month Claude subscription and switch to the free, open-source Agent Harness.
This tool is called DeepSeek Harness. Just two days after it went online, the number of GitHub stars exploded to 99,000. It adopts the MIT license, and its functions are all extended by plug-ins. You can run the free model directly without linking a credit card. Save you $200 a month and keep it safe.
You have reached the end of the archive
All of deepseek